xref: /openbmc/linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 4a0a1436)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /*                                                                        */
4 /*  IBM System i and System p Virtual NIC Device Driver                   */
5 /*  Copyright (C) 2014 IBM Corp.                                          */
6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
9 /*                                                                        */
10 /*                                                                        */
11 /* This module contains the implementation of a virtual ethernet device   */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
15 /*									   */
16 /* Messages are passed between the VNIC driver and the VNIC server using  */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
20 /* are used by the driver to notify the server that a packet is           */
21 /* ready for transmission or that a buffer has been added to receive a    */
22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
23 /* driver that a packet transmission has been completed or that a packet  */
24 /* has been received and placed in a waiting buffer.                      */
25 /*                                                                        */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
28 /* or receive has been completed, the VNIC driver is required to use      */
29 /* "long term mapping". This entails that large, continuous DMA mapped    */
30 /* buffers are allocated on driver initialization and these buffers are   */
31 /* then continuously reused to pass skbs to and from the VNIC server.     */
32 /*                                                                        */
33 /**************************************************************************/
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/kthread.h>
57 #include <linux/seq_file.h>
58 #include <linux/interrupt.h>
59 #include <net/net_namespace.h>
60 #include <asm/hvcall.h>
61 #include <linux/atomic.h>
62 #include <asm/vio.h>
63 #include <asm/iommu.h>
64 #include <linux/uaccess.h>
65 #include <asm/firmware.h>
66 #include <linux/workqueue.h>
67 #include <linux/if_vlan.h>
68 #include <linux/utsname.h>
69 
70 #include "ibmvnic.h"
71 
72 static const char ibmvnic_driver_name[] = "ibmvnic";
73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
74 
75 MODULE_AUTHOR("Santiago Leon");
76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
77 MODULE_LICENSE("GPL");
78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
79 
80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
88 static int enable_scrq_irq(struct ibmvnic_adapter *,
89 			   struct ibmvnic_sub_crq_queue *);
90 static int disable_scrq_irq(struct ibmvnic_adapter *,
91 			    struct ibmvnic_sub_crq_queue *);
92 static int pending_scrq(struct ibmvnic_adapter *,
93 			struct ibmvnic_sub_crq_queue *);
94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
95 					struct ibmvnic_sub_crq_queue *);
96 static int ibmvnic_poll(struct napi_struct *napi, int data);
97 static void send_query_map(struct ibmvnic_adapter *adapter);
98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
100 static int send_login(struct ibmvnic_adapter *adapter);
101 static void send_query_cap(struct ibmvnic_adapter *adapter);
102 static int init_sub_crqs(struct ibmvnic_adapter *);
103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
105 static void release_crq_queue(struct ibmvnic_adapter *);
106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
109 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
110 					 struct ibmvnic_sub_crq_queue *tx_scrq);
111 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
112 				struct ibmvnic_long_term_buff *ltb);
113 
114 struct ibmvnic_stat {
115 	char name[ETH_GSTRING_LEN];
116 	int offset;
117 };
118 
119 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
120 			     offsetof(struct ibmvnic_statistics, stat))
121 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
122 
123 static const struct ibmvnic_stat ibmvnic_stats[] = {
124 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
125 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
126 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
127 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
128 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
129 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
130 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
131 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
132 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
133 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
134 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
135 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
136 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
137 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
138 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
139 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
140 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
141 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
142 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
143 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
144 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
145 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
146 };
147 
148 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
149 {
150 	union ibmvnic_crq crq;
151 
152 	memset(&crq, 0, sizeof(crq));
153 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
154 	crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
155 
156 	return ibmvnic_send_crq(adapter, &crq);
157 }
158 
159 static int send_version_xchg(struct ibmvnic_adapter *adapter)
160 {
161 	union ibmvnic_crq crq;
162 
163 	memset(&crq, 0, sizeof(crq));
164 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
165 	crq.version_exchange.cmd = VERSION_EXCHANGE;
166 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
167 
168 	return ibmvnic_send_crq(adapter, &crq);
169 }
170 
171 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
172 			  unsigned long length, unsigned long *number,
173 			  unsigned long *irq)
174 {
175 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
176 	long rc;
177 
178 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
179 	*number = retbuf[0];
180 	*irq = retbuf[1];
181 
182 	return rc;
183 }
184 
185 /**
186  * ibmvnic_wait_for_completion - Check device state and wait for completion
187  * @adapter: private device data
188  * @comp_done: completion structure to wait for
189  * @timeout: time to wait in milliseconds
190  *
191  * Wait for a completion signal or until the timeout limit is reached
192  * while checking that the device is still active.
193  */
194 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
195 				       struct completion *comp_done,
196 				       unsigned long timeout)
197 {
198 	struct net_device *netdev;
199 	unsigned long div_timeout;
200 	u8 retry;
201 
202 	netdev = adapter->netdev;
203 	retry = 5;
204 	div_timeout = msecs_to_jiffies(timeout / retry);
205 	while (true) {
206 		if (!adapter->crq.active) {
207 			netdev_err(netdev, "Device down!\n");
208 			return -ENODEV;
209 		}
210 		if (!retry--)
211 			break;
212 		if (wait_for_completion_timeout(comp_done, div_timeout))
213 			return 0;
214 	}
215 	netdev_err(netdev, "Operation timed out.\n");
216 	return -ETIMEDOUT;
217 }
218 
219 /**
220  * reuse_ltb() - Check if a long term buffer can be reused
221  * @ltb:  The long term buffer to be checked
222  * @size: The size of the long term buffer.
223  *
224  * An LTB can be reused unless its size has changed.
225  *
226  * Return: Return true if the LTB can be reused, false otherwise.
227  */
228 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
229 {
230 	return (ltb->buff && ltb->size == size);
231 }
232 
233 /**
234  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
235  *
236  * @adapter: ibmvnic adapter associated to the LTB
237  * @ltb:     container object for the LTB
238  * @size:    size of the LTB
239  *
240  * Allocate an LTB of the specified size and notify VIOS.
241  *
242  * If the given @ltb already has the correct size, reuse it. Otherwise if
243  * its non-NULL, free it. Then allocate a new one of the correct size.
244  * Notify the VIOS either way since we may now be working with a new VIOS.
245  *
246  * Allocating larger chunks of memory during resets, specially LPM or under
247  * low memory situations can cause resets to fail/timeout and for LPAR to
248  * lose connectivity. So hold onto the LTB even if we fail to communicate
249  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
250  *
251  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
252  *	   a negative value otherwise.
253  */
254 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
255 				struct ibmvnic_long_term_buff *ltb, int size)
256 {
257 	struct device *dev = &adapter->vdev->dev;
258 	int rc;
259 
260 	if (!reuse_ltb(ltb, size)) {
261 		dev_dbg(dev,
262 			"LTB size changed from 0x%llx to 0x%x, reallocating\n",
263 			 ltb->size, size);
264 		free_long_term_buff(adapter, ltb);
265 	}
266 
267 	if (ltb->buff) {
268 		dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
269 			ltb->map_id, ltb->size);
270 	} else {
271 		ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
272 					       GFP_KERNEL);
273 		if (!ltb->buff) {
274 			dev_err(dev, "Couldn't alloc long term buffer\n");
275 			return -ENOMEM;
276 		}
277 		ltb->size = size;
278 
279 		ltb->map_id = find_first_zero_bit(adapter->map_ids,
280 						  MAX_MAP_ID);
281 		bitmap_set(adapter->map_ids, ltb->map_id, 1);
282 
283 		dev_dbg(dev,
284 			"Allocated new LTB [map %d, size 0x%llx]\n",
285 			 ltb->map_id, ltb->size);
286 	}
287 
288 	/* Ensure ltb is zeroed - specially when reusing it. */
289 	memset(ltb->buff, 0, ltb->size);
290 
291 	mutex_lock(&adapter->fw_lock);
292 	adapter->fw_done_rc = 0;
293 	reinit_completion(&adapter->fw_done);
294 
295 	rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
296 	if (rc) {
297 		dev_err(dev, "send_request_map failed, rc = %d\n", rc);
298 		goto out;
299 	}
300 
301 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
302 	if (rc) {
303 		dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
304 			rc);
305 		goto out;
306 	}
307 
308 	if (adapter->fw_done_rc) {
309 		dev_err(dev, "Couldn't map LTB, rc = %d\n",
310 			adapter->fw_done_rc);
311 		rc = -EIO;
312 		goto out;
313 	}
314 	rc = 0;
315 out:
316 	/* don't free LTB on communication error - see function header */
317 	mutex_unlock(&adapter->fw_lock);
318 	return rc;
319 }
320 
321 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
322 				struct ibmvnic_long_term_buff *ltb)
323 {
324 	struct device *dev = &adapter->vdev->dev;
325 
326 	if (!ltb->buff)
327 		return;
328 
329 	/* VIOS automatically unmaps the long term buffer at remote
330 	 * end for the following resets:
331 	 * FAILOVER, MOBILITY, TIMEOUT.
332 	 */
333 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
334 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
335 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
336 		send_request_unmap(adapter, ltb->map_id);
337 
338 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
339 
340 	ltb->buff = NULL;
341 	/* mark this map_id free */
342 	bitmap_clear(adapter->map_ids, ltb->map_id, 1);
343 	ltb->map_id = 0;
344 }
345 
346 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
347 {
348 	int i;
349 
350 	for (i = 0; i < adapter->num_active_rx_pools; i++)
351 		adapter->rx_pool[i].active = 0;
352 }
353 
354 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
355 			      struct ibmvnic_rx_pool *pool)
356 {
357 	int count = pool->size - atomic_read(&pool->available);
358 	u64 handle = adapter->rx_scrq[pool->index]->handle;
359 	struct device *dev = &adapter->vdev->dev;
360 	struct ibmvnic_ind_xmit_queue *ind_bufp;
361 	struct ibmvnic_sub_crq_queue *rx_scrq;
362 	union sub_crq *sub_crq;
363 	int buffers_added = 0;
364 	unsigned long lpar_rc;
365 	struct sk_buff *skb;
366 	unsigned int offset;
367 	dma_addr_t dma_addr;
368 	unsigned char *dst;
369 	int shift = 0;
370 	int index;
371 	int i;
372 
373 	if (!pool->active)
374 		return;
375 
376 	rx_scrq = adapter->rx_scrq[pool->index];
377 	ind_bufp = &rx_scrq->ind_buf;
378 
379 	/* netdev_skb_alloc() could have failed after we saved a few skbs
380 	 * in the indir_buf and we would not have sent them to VIOS yet.
381 	 * To account for them, start the loop at ind_bufp->index rather
382 	 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
383 	 * be 0.
384 	 */
385 	for (i = ind_bufp->index; i < count; ++i) {
386 		index = pool->free_map[pool->next_free];
387 
388 		/* We maybe reusing the skb from earlier resets. Allocate
389 		 * only if necessary. But since the LTB may have changed
390 		 * during reset (see init_rx_pools()), update LTB below
391 		 * even if reusing skb.
392 		 */
393 		skb = pool->rx_buff[index].skb;
394 		if (!skb) {
395 			skb = netdev_alloc_skb(adapter->netdev,
396 					       pool->buff_size);
397 			if (!skb) {
398 				dev_err(dev, "Couldn't replenish rx buff\n");
399 				adapter->replenish_no_mem++;
400 				break;
401 			}
402 		}
403 
404 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
405 		pool->next_free = (pool->next_free + 1) % pool->size;
406 
407 		/* Copy the skb to the long term mapped DMA buffer */
408 		offset = index * pool->buff_size;
409 		dst = pool->long_term_buff.buff + offset;
410 		memset(dst, 0, pool->buff_size);
411 		dma_addr = pool->long_term_buff.addr + offset;
412 
413 		/* add the skb to an rx_buff in the pool */
414 		pool->rx_buff[index].data = dst;
415 		pool->rx_buff[index].dma = dma_addr;
416 		pool->rx_buff[index].skb = skb;
417 		pool->rx_buff[index].pool_index = pool->index;
418 		pool->rx_buff[index].size = pool->buff_size;
419 
420 		/* queue the rx_buff for the next send_subcrq_indirect */
421 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
422 		memset(sub_crq, 0, sizeof(*sub_crq));
423 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
424 		sub_crq->rx_add.correlator =
425 		    cpu_to_be64((u64)&pool->rx_buff[index]);
426 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
427 		sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
428 
429 		/* The length field of the sCRQ is defined to be 24 bits so the
430 		 * buffer size needs to be left shifted by a byte before it is
431 		 * converted to big endian to prevent the last byte from being
432 		 * truncated.
433 		 */
434 #ifdef __LITTLE_ENDIAN__
435 		shift = 8;
436 #endif
437 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
438 
439 		/* if send_subcrq_indirect queue is full, flush to VIOS */
440 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
441 		    i == count - 1) {
442 			lpar_rc =
443 				send_subcrq_indirect(adapter, handle,
444 						     (u64)ind_bufp->indir_dma,
445 						     (u64)ind_bufp->index);
446 			if (lpar_rc != H_SUCCESS)
447 				goto failure;
448 			buffers_added += ind_bufp->index;
449 			adapter->replenish_add_buff_success += ind_bufp->index;
450 			ind_bufp->index = 0;
451 		}
452 	}
453 	atomic_add(buffers_added, &pool->available);
454 	return;
455 
456 failure:
457 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
458 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
459 	for (i = ind_bufp->index - 1; i >= 0; --i) {
460 		struct ibmvnic_rx_buff *rx_buff;
461 
462 		pool->next_free = pool->next_free == 0 ?
463 				  pool->size - 1 : pool->next_free - 1;
464 		sub_crq = &ind_bufp->indir_arr[i];
465 		rx_buff = (struct ibmvnic_rx_buff *)
466 				be64_to_cpu(sub_crq->rx_add.correlator);
467 		index = (int)(rx_buff - pool->rx_buff);
468 		pool->free_map[pool->next_free] = index;
469 		dev_kfree_skb_any(pool->rx_buff[index].skb);
470 		pool->rx_buff[index].skb = NULL;
471 	}
472 	adapter->replenish_add_buff_failure += ind_bufp->index;
473 	atomic_add(buffers_added, &pool->available);
474 	ind_bufp->index = 0;
475 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
476 		/* Disable buffer pool replenishment and report carrier off if
477 		 * queue is closed or pending failover.
478 		 * Firmware guarantees that a signal will be sent to the
479 		 * driver, triggering a reset.
480 		 */
481 		deactivate_rx_pools(adapter);
482 		netif_carrier_off(adapter->netdev);
483 	}
484 }
485 
486 static void replenish_pools(struct ibmvnic_adapter *adapter)
487 {
488 	int i;
489 
490 	adapter->replenish_task_cycles++;
491 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
492 		if (adapter->rx_pool[i].active)
493 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
494 	}
495 
496 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
497 }
498 
499 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
500 {
501 	kfree(adapter->tx_stats_buffers);
502 	kfree(adapter->rx_stats_buffers);
503 	adapter->tx_stats_buffers = NULL;
504 	adapter->rx_stats_buffers = NULL;
505 }
506 
507 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
508 {
509 	adapter->tx_stats_buffers =
510 				kcalloc(IBMVNIC_MAX_QUEUES,
511 					sizeof(struct ibmvnic_tx_queue_stats),
512 					GFP_KERNEL);
513 	if (!adapter->tx_stats_buffers)
514 		return -ENOMEM;
515 
516 	adapter->rx_stats_buffers =
517 				kcalloc(IBMVNIC_MAX_QUEUES,
518 					sizeof(struct ibmvnic_rx_queue_stats),
519 					GFP_KERNEL);
520 	if (!adapter->rx_stats_buffers)
521 		return -ENOMEM;
522 
523 	return 0;
524 }
525 
526 static void release_stats_token(struct ibmvnic_adapter *adapter)
527 {
528 	struct device *dev = &adapter->vdev->dev;
529 
530 	if (!adapter->stats_token)
531 		return;
532 
533 	dma_unmap_single(dev, adapter->stats_token,
534 			 sizeof(struct ibmvnic_statistics),
535 			 DMA_FROM_DEVICE);
536 	adapter->stats_token = 0;
537 }
538 
539 static int init_stats_token(struct ibmvnic_adapter *adapter)
540 {
541 	struct device *dev = &adapter->vdev->dev;
542 	dma_addr_t stok;
543 	int rc;
544 
545 	stok = dma_map_single(dev, &adapter->stats,
546 			      sizeof(struct ibmvnic_statistics),
547 			      DMA_FROM_DEVICE);
548 	rc = dma_mapping_error(dev, stok);
549 	if (rc) {
550 		dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
551 		return rc;
552 	}
553 
554 	adapter->stats_token = stok;
555 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
556 	return 0;
557 }
558 
559 /**
560  * release_rx_pools() - Release any rx pools attached to @adapter.
561  * @adapter: ibmvnic adapter
562  *
563  * Safe to call this multiple times - even if no pools are attached.
564  */
565 static void release_rx_pools(struct ibmvnic_adapter *adapter)
566 {
567 	struct ibmvnic_rx_pool *rx_pool;
568 	int i, j;
569 
570 	if (!adapter->rx_pool)
571 		return;
572 
573 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
574 		rx_pool = &adapter->rx_pool[i];
575 
576 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
577 
578 		kfree(rx_pool->free_map);
579 
580 		free_long_term_buff(adapter, &rx_pool->long_term_buff);
581 
582 		if (!rx_pool->rx_buff)
583 			continue;
584 
585 		for (j = 0; j < rx_pool->size; j++) {
586 			if (rx_pool->rx_buff[j].skb) {
587 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
588 				rx_pool->rx_buff[j].skb = NULL;
589 			}
590 		}
591 
592 		kfree(rx_pool->rx_buff);
593 	}
594 
595 	kfree(adapter->rx_pool);
596 	adapter->rx_pool = NULL;
597 	adapter->num_active_rx_pools = 0;
598 	adapter->prev_rx_pool_size = 0;
599 }
600 
601 /**
602  * reuse_rx_pools() - Check if the existing rx pools can be reused.
603  * @adapter: ibmvnic adapter
604  *
605  * Check if the existing rx pools in the adapter can be reused. The
606  * pools can be reused if the pool parameters (number of pools,
607  * number of buffers in the pool and size of each buffer) have not
608  * changed.
609  *
610  * NOTE: This assumes that all pools have the same number of buffers
611  *       which is the case currently. If that changes, we must fix this.
612  *
613  * Return: true if the rx pools can be reused, false otherwise.
614  */
615 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
616 {
617 	u64 old_num_pools, new_num_pools;
618 	u64 old_pool_size, new_pool_size;
619 	u64 old_buff_size, new_buff_size;
620 
621 	if (!adapter->rx_pool)
622 		return false;
623 
624 	old_num_pools = adapter->num_active_rx_pools;
625 	new_num_pools = adapter->req_rx_queues;
626 
627 	old_pool_size = adapter->prev_rx_pool_size;
628 	new_pool_size = adapter->req_rx_add_entries_per_subcrq;
629 
630 	old_buff_size = adapter->prev_rx_buf_sz;
631 	new_buff_size = adapter->cur_rx_buf_sz;
632 
633 	if (old_buff_size != new_buff_size ||
634 	    old_num_pools != new_num_pools ||
635 	    old_pool_size != new_pool_size)
636 		return false;
637 
638 	return true;
639 }
640 
641 /**
642  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
643  * @netdev: net device associated with the vnic interface
644  *
645  * Initialize the set of receiver pools in the ibmvnic adapter associated
646  * with the net_device @netdev. If possible, reuse the existing rx pools.
647  * Otherwise free any existing pools and  allocate a new set of pools
648  * before initializing them.
649  *
650  * Return: 0 on success and negative value on error.
651  */
652 static int init_rx_pools(struct net_device *netdev)
653 {
654 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
655 	struct device *dev = &adapter->vdev->dev;
656 	struct ibmvnic_rx_pool *rx_pool;
657 	u64 num_pools;
658 	u64 pool_size;		/* # of buffers in one pool */
659 	u64 buff_size;
660 	int i, j, rc;
661 
662 	pool_size = adapter->req_rx_add_entries_per_subcrq;
663 	num_pools = adapter->req_rx_queues;
664 	buff_size = adapter->cur_rx_buf_sz;
665 
666 	if (reuse_rx_pools(adapter)) {
667 		dev_dbg(dev, "Reusing rx pools\n");
668 		goto update_ltb;
669 	}
670 
671 	/* Allocate/populate the pools. */
672 	release_rx_pools(adapter);
673 
674 	adapter->rx_pool = kcalloc(num_pools,
675 				   sizeof(struct ibmvnic_rx_pool),
676 				   GFP_KERNEL);
677 	if (!adapter->rx_pool) {
678 		dev_err(dev, "Failed to allocate rx pools\n");
679 		return -ENOMEM;
680 	}
681 
682 	/* Set num_active_rx_pools early. If we fail below after partial
683 	 * allocation, release_rx_pools() will know how many to look for.
684 	 */
685 	adapter->num_active_rx_pools = num_pools;
686 
687 	for (i = 0; i < num_pools; i++) {
688 		rx_pool = &adapter->rx_pool[i];
689 
690 		netdev_dbg(adapter->netdev,
691 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
692 			   i, pool_size, buff_size);
693 
694 		rx_pool->size = pool_size;
695 		rx_pool->index = i;
696 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
697 
698 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
699 					    GFP_KERNEL);
700 		if (!rx_pool->free_map) {
701 			dev_err(dev, "Couldn't alloc free_map %d\n", i);
702 			rc = -ENOMEM;
703 			goto out_release;
704 		}
705 
706 		rx_pool->rx_buff = kcalloc(rx_pool->size,
707 					   sizeof(struct ibmvnic_rx_buff),
708 					   GFP_KERNEL);
709 		if (!rx_pool->rx_buff) {
710 			dev_err(dev, "Couldn't alloc rx buffers\n");
711 			rc = -ENOMEM;
712 			goto out_release;
713 		}
714 	}
715 
716 	adapter->prev_rx_pool_size = pool_size;
717 	adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
718 
719 update_ltb:
720 	for (i = 0; i < num_pools; i++) {
721 		rx_pool = &adapter->rx_pool[i];
722 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
723 			i, rx_pool->size, rx_pool->buff_size);
724 
725 		rc = alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
726 					  rx_pool->size * rx_pool->buff_size);
727 		if (rc)
728 			goto out;
729 
730 		for (j = 0; j < rx_pool->size; ++j) {
731 			struct ibmvnic_rx_buff *rx_buff;
732 
733 			rx_pool->free_map[j] = j;
734 
735 			/* NOTE: Don't clear rx_buff->skb here - will leak
736 			 * memory! replenish_rx_pool() will reuse skbs or
737 			 * allocate as necessary.
738 			 */
739 			rx_buff = &rx_pool->rx_buff[j];
740 			rx_buff->dma = 0;
741 			rx_buff->data = 0;
742 			rx_buff->size = 0;
743 			rx_buff->pool_index = 0;
744 		}
745 
746 		/* Mark pool "empty" so replenish_rx_pools() will
747 		 * update the LTB info for each buffer
748 		 */
749 		atomic_set(&rx_pool->available, 0);
750 		rx_pool->next_alloc = 0;
751 		rx_pool->next_free = 0;
752 		/* replenish_rx_pool() may have called deactivate_rx_pools()
753 		 * on failover. Ensure pool is active now.
754 		 */
755 		rx_pool->active = 1;
756 	}
757 	return 0;
758 out_release:
759 	release_rx_pools(adapter);
760 out:
761 	/* We failed to allocate one or more LTBs or map them on the VIOS.
762 	 * Hold onto the pools and any LTBs that we did allocate/map.
763 	 */
764 	return rc;
765 }
766 
767 static void release_vpd_data(struct ibmvnic_adapter *adapter)
768 {
769 	if (!adapter->vpd)
770 		return;
771 
772 	kfree(adapter->vpd->buff);
773 	kfree(adapter->vpd);
774 
775 	adapter->vpd = NULL;
776 }
777 
778 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
779 				struct ibmvnic_tx_pool *tx_pool)
780 {
781 	kfree(tx_pool->tx_buff);
782 	kfree(tx_pool->free_map);
783 	free_long_term_buff(adapter, &tx_pool->long_term_buff);
784 }
785 
786 /**
787  * release_tx_pools() - Release any tx pools attached to @adapter.
788  * @adapter: ibmvnic adapter
789  *
790  * Safe to call this multiple times - even if no pools are attached.
791  */
792 static void release_tx_pools(struct ibmvnic_adapter *adapter)
793 {
794 	int i;
795 
796 	/* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
797 	 * both NULL or both non-NULL. So we only need to check one.
798 	 */
799 	if (!adapter->tx_pool)
800 		return;
801 
802 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
803 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
804 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
805 	}
806 
807 	kfree(adapter->tx_pool);
808 	adapter->tx_pool = NULL;
809 	kfree(adapter->tso_pool);
810 	adapter->tso_pool = NULL;
811 	adapter->num_active_tx_pools = 0;
812 	adapter->prev_tx_pool_size = 0;
813 }
814 
815 static int init_one_tx_pool(struct net_device *netdev,
816 			    struct ibmvnic_tx_pool *tx_pool,
817 			    int pool_size, int buf_size)
818 {
819 	int i;
820 
821 	tx_pool->tx_buff = kcalloc(pool_size,
822 				   sizeof(struct ibmvnic_tx_buff),
823 				   GFP_KERNEL);
824 	if (!tx_pool->tx_buff)
825 		return -ENOMEM;
826 
827 	tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
828 	if (!tx_pool->free_map) {
829 		kfree(tx_pool->tx_buff);
830 		tx_pool->tx_buff = NULL;
831 		return -ENOMEM;
832 	}
833 
834 	for (i = 0; i < pool_size; i++)
835 		tx_pool->free_map[i] = i;
836 
837 	tx_pool->consumer_index = 0;
838 	tx_pool->producer_index = 0;
839 	tx_pool->num_buffers = pool_size;
840 	tx_pool->buf_size = buf_size;
841 
842 	return 0;
843 }
844 
845 /**
846  * reuse_tx_pools() - Check if the existing tx pools can be reused.
847  * @adapter: ibmvnic adapter
848  *
849  * Check if the existing tx pools in the adapter can be reused. The
850  * pools can be reused if the pool parameters (number of pools,
851  * number of buffers in the pool and mtu) have not changed.
852  *
853  * NOTE: This assumes that all pools have the same number of buffers
854  *       which is the case currently. If that changes, we must fix this.
855  *
856  * Return: true if the tx pools can be reused, false otherwise.
857  */
858 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
859 {
860 	u64 old_num_pools, new_num_pools;
861 	u64 old_pool_size, new_pool_size;
862 	u64 old_mtu, new_mtu;
863 
864 	if (!adapter->tx_pool)
865 		return false;
866 
867 	old_num_pools = adapter->num_active_tx_pools;
868 	new_num_pools = adapter->num_active_tx_scrqs;
869 	old_pool_size = adapter->prev_tx_pool_size;
870 	new_pool_size = adapter->req_tx_entries_per_subcrq;
871 	old_mtu = adapter->prev_mtu;
872 	new_mtu = adapter->req_mtu;
873 
874 	if (old_mtu != new_mtu ||
875 	    old_num_pools != new_num_pools ||
876 	    old_pool_size != new_pool_size)
877 		return false;
878 
879 	return true;
880 }
881 
882 /**
883  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
884  * @netdev: net device associated with the vnic interface
885  *
886  * Initialize the set of transmit pools in the ibmvnic adapter associated
887  * with the net_device @netdev. If possible, reuse the existing tx pools.
888  * Otherwise free any existing pools and  allocate a new set of pools
889  * before initializing them.
890  *
891  * Return: 0 on success and negative value on error.
892  */
893 static int init_tx_pools(struct net_device *netdev)
894 {
895 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
896 	struct device *dev = &adapter->vdev->dev;
897 	int num_pools;
898 	u64 pool_size;		/* # of buffers in pool */
899 	u64 buff_size;
900 	int i, j, rc;
901 
902 	num_pools = adapter->req_tx_queues;
903 
904 	/* We must notify the VIOS about the LTB on all resets - but we only
905 	 * need to alloc/populate pools if either the number of buffers or
906 	 * size of each buffer in the pool has changed.
907 	 */
908 	if (reuse_tx_pools(adapter)) {
909 		netdev_dbg(netdev, "Reusing tx pools\n");
910 		goto update_ltb;
911 	}
912 
913 	/* Allocate/populate the pools. */
914 	release_tx_pools(adapter);
915 
916 	pool_size = adapter->req_tx_entries_per_subcrq;
917 	num_pools = adapter->num_active_tx_scrqs;
918 
919 	adapter->tx_pool = kcalloc(num_pools,
920 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
921 	if (!adapter->tx_pool)
922 		return -ENOMEM;
923 
924 	adapter->tso_pool = kcalloc(num_pools,
925 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
926 	/* To simplify release_tx_pools() ensure that ->tx_pool and
927 	 * ->tso_pool are either both NULL or both non-NULL.
928 	 */
929 	if (!adapter->tso_pool) {
930 		kfree(adapter->tx_pool);
931 		adapter->tx_pool = NULL;
932 		return -ENOMEM;
933 	}
934 
935 	/* Set num_active_tx_pools early. If we fail below after partial
936 	 * allocation, release_tx_pools() will know how many to look for.
937 	 */
938 	adapter->num_active_tx_pools = num_pools;
939 
940 	buff_size = adapter->req_mtu + VLAN_HLEN;
941 	buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
942 
943 	for (i = 0; i < num_pools; i++) {
944 		dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
945 			i, adapter->req_tx_entries_per_subcrq, buff_size);
946 
947 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
948 				      pool_size, buff_size);
949 		if (rc)
950 			goto out_release;
951 
952 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
953 				      IBMVNIC_TSO_BUFS,
954 				      IBMVNIC_TSO_BUF_SZ);
955 		if (rc)
956 			goto out_release;
957 	}
958 
959 	adapter->prev_tx_pool_size = pool_size;
960 	adapter->prev_mtu = adapter->req_mtu;
961 
962 update_ltb:
963 	/* NOTE: All tx_pools have the same number of buffers (which is
964 	 *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
965 	 *       buffers (see calls init_one_tx_pool() for these).
966 	 *       For consistency, we use tx_pool->num_buffers and
967 	 *       tso_pool->num_buffers below.
968 	 */
969 	rc = -1;
970 	for (i = 0; i < num_pools; i++) {
971 		struct ibmvnic_tx_pool *tso_pool;
972 		struct ibmvnic_tx_pool *tx_pool;
973 		u32 ltb_size;
974 
975 		tx_pool = &adapter->tx_pool[i];
976 		ltb_size = tx_pool->num_buffers * tx_pool->buf_size;
977 		if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
978 					 ltb_size))
979 			goto out;
980 
981 		dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n",
982 			i, tx_pool->long_term_buff.buff,
983 			tx_pool->num_buffers, tx_pool->buf_size);
984 
985 		tx_pool->consumer_index = 0;
986 		tx_pool->producer_index = 0;
987 
988 		for (j = 0; j < tx_pool->num_buffers; j++)
989 			tx_pool->free_map[j] = j;
990 
991 		tso_pool = &adapter->tso_pool[i];
992 		ltb_size = tso_pool->num_buffers * tso_pool->buf_size;
993 		if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff,
994 					 ltb_size))
995 			goto out;
996 
997 		dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n",
998 			i, tso_pool->long_term_buff.buff,
999 			tso_pool->num_buffers, tso_pool->buf_size);
1000 
1001 		tso_pool->consumer_index = 0;
1002 		tso_pool->producer_index = 0;
1003 
1004 		for (j = 0; j < tso_pool->num_buffers; j++)
1005 			tso_pool->free_map[j] = j;
1006 	}
1007 
1008 	return 0;
1009 out_release:
1010 	release_tx_pools(adapter);
1011 out:
1012 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1013 	 * Hold onto the pools and any LTBs that we did allocate/map.
1014 	 */
1015 	return rc;
1016 }
1017 
1018 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1019 {
1020 	int i;
1021 
1022 	if (adapter->napi_enabled)
1023 		return;
1024 
1025 	for (i = 0; i < adapter->req_rx_queues; i++)
1026 		napi_enable(&adapter->napi[i]);
1027 
1028 	adapter->napi_enabled = true;
1029 }
1030 
1031 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1032 {
1033 	int i;
1034 
1035 	if (!adapter->napi_enabled)
1036 		return;
1037 
1038 	for (i = 0; i < adapter->req_rx_queues; i++) {
1039 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1040 		napi_disable(&adapter->napi[i]);
1041 	}
1042 
1043 	adapter->napi_enabled = false;
1044 }
1045 
1046 static int init_napi(struct ibmvnic_adapter *adapter)
1047 {
1048 	int i;
1049 
1050 	adapter->napi = kcalloc(adapter->req_rx_queues,
1051 				sizeof(struct napi_struct), GFP_KERNEL);
1052 	if (!adapter->napi)
1053 		return -ENOMEM;
1054 
1055 	for (i = 0; i < adapter->req_rx_queues; i++) {
1056 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1057 		netif_napi_add(adapter->netdev, &adapter->napi[i],
1058 			       ibmvnic_poll, NAPI_POLL_WEIGHT);
1059 	}
1060 
1061 	adapter->num_active_rx_napi = adapter->req_rx_queues;
1062 	return 0;
1063 }
1064 
1065 static void release_napi(struct ibmvnic_adapter *adapter)
1066 {
1067 	int i;
1068 
1069 	if (!adapter->napi)
1070 		return;
1071 
1072 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
1073 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1074 		netif_napi_del(&adapter->napi[i]);
1075 	}
1076 
1077 	kfree(adapter->napi);
1078 	adapter->napi = NULL;
1079 	adapter->num_active_rx_napi = 0;
1080 	adapter->napi_enabled = false;
1081 }
1082 
1083 static const char *adapter_state_to_string(enum vnic_state state)
1084 {
1085 	switch (state) {
1086 	case VNIC_PROBING:
1087 		return "PROBING";
1088 	case VNIC_PROBED:
1089 		return "PROBED";
1090 	case VNIC_OPENING:
1091 		return "OPENING";
1092 	case VNIC_OPEN:
1093 		return "OPEN";
1094 	case VNIC_CLOSING:
1095 		return "CLOSING";
1096 	case VNIC_CLOSED:
1097 		return "CLOSED";
1098 	case VNIC_REMOVING:
1099 		return "REMOVING";
1100 	case VNIC_REMOVED:
1101 		return "REMOVED";
1102 	case VNIC_DOWN:
1103 		return "DOWN";
1104 	}
1105 	return "UNKNOWN";
1106 }
1107 
1108 static int ibmvnic_login(struct net_device *netdev)
1109 {
1110 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1111 	unsigned long timeout = msecs_to_jiffies(20000);
1112 	int retry_count = 0;
1113 	int retries = 10;
1114 	bool retry;
1115 	int rc;
1116 
1117 	do {
1118 		retry = false;
1119 		if (retry_count > retries) {
1120 			netdev_warn(netdev, "Login attempts exceeded\n");
1121 			return -EACCES;
1122 		}
1123 
1124 		adapter->init_done_rc = 0;
1125 		reinit_completion(&adapter->init_done);
1126 		rc = send_login(adapter);
1127 		if (rc)
1128 			return rc;
1129 
1130 		if (!wait_for_completion_timeout(&adapter->init_done,
1131 						 timeout)) {
1132 			netdev_warn(netdev, "Login timed out, retrying...\n");
1133 			retry = true;
1134 			adapter->init_done_rc = 0;
1135 			retry_count++;
1136 			continue;
1137 		}
1138 
1139 		if (adapter->init_done_rc == ABORTED) {
1140 			netdev_warn(netdev, "Login aborted, retrying...\n");
1141 			retry = true;
1142 			adapter->init_done_rc = 0;
1143 			retry_count++;
1144 			/* FW or device may be busy, so
1145 			 * wait a bit before retrying login
1146 			 */
1147 			msleep(500);
1148 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
1149 			retry_count++;
1150 			release_sub_crqs(adapter, 1);
1151 
1152 			retry = true;
1153 			netdev_dbg(netdev,
1154 				   "Received partial success, retrying...\n");
1155 			adapter->init_done_rc = 0;
1156 			reinit_completion(&adapter->init_done);
1157 			send_query_cap(adapter);
1158 			if (!wait_for_completion_timeout(&adapter->init_done,
1159 							 timeout)) {
1160 				netdev_warn(netdev,
1161 					    "Capabilities query timed out\n");
1162 				return -ETIMEDOUT;
1163 			}
1164 
1165 			rc = init_sub_crqs(adapter);
1166 			if (rc) {
1167 				netdev_warn(netdev,
1168 					    "SCRQ initialization failed\n");
1169 				return rc;
1170 			}
1171 
1172 			rc = init_sub_crq_irqs(adapter);
1173 			if (rc) {
1174 				netdev_warn(netdev,
1175 					    "SCRQ irq initialization failed\n");
1176 				return rc;
1177 			}
1178 		} else if (adapter->init_done_rc) {
1179 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1180 				    adapter->init_done_rc);
1181 			return -EIO;
1182 		}
1183 	} while (retry);
1184 
1185 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
1186 
1187 	netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1188 	return 0;
1189 }
1190 
1191 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1192 {
1193 	kfree(adapter->login_buf);
1194 	adapter->login_buf = NULL;
1195 }
1196 
1197 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1198 {
1199 	kfree(adapter->login_rsp_buf);
1200 	adapter->login_rsp_buf = NULL;
1201 }
1202 
1203 static void release_resources(struct ibmvnic_adapter *adapter)
1204 {
1205 	release_vpd_data(adapter);
1206 
1207 	release_napi(adapter);
1208 	release_login_buffer(adapter);
1209 	release_login_rsp_buffer(adapter);
1210 }
1211 
1212 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1213 {
1214 	struct net_device *netdev = adapter->netdev;
1215 	unsigned long timeout = msecs_to_jiffies(20000);
1216 	union ibmvnic_crq crq;
1217 	bool resend;
1218 	int rc;
1219 
1220 	netdev_dbg(netdev, "setting link state %d\n", link_state);
1221 
1222 	memset(&crq, 0, sizeof(crq));
1223 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1224 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1225 	crq.logical_link_state.link_state = link_state;
1226 
1227 	do {
1228 		resend = false;
1229 
1230 		reinit_completion(&adapter->init_done);
1231 		rc = ibmvnic_send_crq(adapter, &crq);
1232 		if (rc) {
1233 			netdev_err(netdev, "Failed to set link state\n");
1234 			return rc;
1235 		}
1236 
1237 		if (!wait_for_completion_timeout(&adapter->init_done,
1238 						 timeout)) {
1239 			netdev_err(netdev, "timeout setting link state\n");
1240 			return -ETIMEDOUT;
1241 		}
1242 
1243 		if (adapter->init_done_rc == PARTIALSUCCESS) {
1244 			/* Partuial success, delay and re-send */
1245 			mdelay(1000);
1246 			resend = true;
1247 		} else if (adapter->init_done_rc) {
1248 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1249 				    adapter->init_done_rc);
1250 			return adapter->init_done_rc;
1251 		}
1252 	} while (resend);
1253 
1254 	return 0;
1255 }
1256 
1257 static int set_real_num_queues(struct net_device *netdev)
1258 {
1259 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1260 	int rc;
1261 
1262 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1263 		   adapter->req_tx_queues, adapter->req_rx_queues);
1264 
1265 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1266 	if (rc) {
1267 		netdev_err(netdev, "failed to set the number of tx queues\n");
1268 		return rc;
1269 	}
1270 
1271 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1272 	if (rc)
1273 		netdev_err(netdev, "failed to set the number of rx queues\n");
1274 
1275 	return rc;
1276 }
1277 
1278 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1279 {
1280 	struct device *dev = &adapter->vdev->dev;
1281 	union ibmvnic_crq crq;
1282 	int len = 0;
1283 	int rc;
1284 
1285 	if (adapter->vpd->buff)
1286 		len = adapter->vpd->len;
1287 
1288 	mutex_lock(&adapter->fw_lock);
1289 	adapter->fw_done_rc = 0;
1290 	reinit_completion(&adapter->fw_done);
1291 
1292 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1293 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1294 	rc = ibmvnic_send_crq(adapter, &crq);
1295 	if (rc) {
1296 		mutex_unlock(&adapter->fw_lock);
1297 		return rc;
1298 	}
1299 
1300 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1301 	if (rc) {
1302 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1303 		mutex_unlock(&adapter->fw_lock);
1304 		return rc;
1305 	}
1306 	mutex_unlock(&adapter->fw_lock);
1307 
1308 	if (!adapter->vpd->len)
1309 		return -ENODATA;
1310 
1311 	if (!adapter->vpd->buff)
1312 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1313 	else if (adapter->vpd->len != len)
1314 		adapter->vpd->buff =
1315 			krealloc(adapter->vpd->buff,
1316 				 adapter->vpd->len, GFP_KERNEL);
1317 
1318 	if (!adapter->vpd->buff) {
1319 		dev_err(dev, "Could allocate VPD buffer\n");
1320 		return -ENOMEM;
1321 	}
1322 
1323 	adapter->vpd->dma_addr =
1324 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1325 			       DMA_FROM_DEVICE);
1326 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1327 		dev_err(dev, "Could not map VPD buffer\n");
1328 		kfree(adapter->vpd->buff);
1329 		adapter->vpd->buff = NULL;
1330 		return -ENOMEM;
1331 	}
1332 
1333 	mutex_lock(&adapter->fw_lock);
1334 	adapter->fw_done_rc = 0;
1335 	reinit_completion(&adapter->fw_done);
1336 
1337 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1338 	crq.get_vpd.cmd = GET_VPD;
1339 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1340 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1341 	rc = ibmvnic_send_crq(adapter, &crq);
1342 	if (rc) {
1343 		kfree(adapter->vpd->buff);
1344 		adapter->vpd->buff = NULL;
1345 		mutex_unlock(&adapter->fw_lock);
1346 		return rc;
1347 	}
1348 
1349 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1350 	if (rc) {
1351 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1352 		kfree(adapter->vpd->buff);
1353 		adapter->vpd->buff = NULL;
1354 		mutex_unlock(&adapter->fw_lock);
1355 		return rc;
1356 	}
1357 
1358 	mutex_unlock(&adapter->fw_lock);
1359 	return 0;
1360 }
1361 
1362 static int init_resources(struct ibmvnic_adapter *adapter)
1363 {
1364 	struct net_device *netdev = adapter->netdev;
1365 	int rc;
1366 
1367 	rc = set_real_num_queues(netdev);
1368 	if (rc)
1369 		return rc;
1370 
1371 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1372 	if (!adapter->vpd)
1373 		return -ENOMEM;
1374 
1375 	/* Vital Product Data (VPD) */
1376 	rc = ibmvnic_get_vpd(adapter);
1377 	if (rc) {
1378 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1379 		return rc;
1380 	}
1381 
1382 	rc = init_napi(adapter);
1383 	if (rc)
1384 		return rc;
1385 
1386 	send_query_map(adapter);
1387 
1388 	rc = init_rx_pools(netdev);
1389 	if (rc)
1390 		return rc;
1391 
1392 	rc = init_tx_pools(netdev);
1393 	return rc;
1394 }
1395 
1396 static int __ibmvnic_open(struct net_device *netdev)
1397 {
1398 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1399 	enum vnic_state prev_state = adapter->state;
1400 	int i, rc;
1401 
1402 	adapter->state = VNIC_OPENING;
1403 	replenish_pools(adapter);
1404 	ibmvnic_napi_enable(adapter);
1405 
1406 	/* We're ready to receive frames, enable the sub-crq interrupts and
1407 	 * set the logical link state to up
1408 	 */
1409 	for (i = 0; i < adapter->req_rx_queues; i++) {
1410 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1411 		if (prev_state == VNIC_CLOSED)
1412 			enable_irq(adapter->rx_scrq[i]->irq);
1413 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1414 	}
1415 
1416 	for (i = 0; i < adapter->req_tx_queues; i++) {
1417 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1418 		if (prev_state == VNIC_CLOSED)
1419 			enable_irq(adapter->tx_scrq[i]->irq);
1420 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1421 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1422 	}
1423 
1424 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1425 	if (rc) {
1426 		ibmvnic_napi_disable(adapter);
1427 		release_resources(adapter);
1428 		return rc;
1429 	}
1430 
1431 	netif_tx_start_all_queues(netdev);
1432 
1433 	if (prev_state == VNIC_CLOSED) {
1434 		for (i = 0; i < adapter->req_rx_queues; i++)
1435 			napi_schedule(&adapter->napi[i]);
1436 	}
1437 
1438 	adapter->state = VNIC_OPEN;
1439 	return rc;
1440 }
1441 
1442 static int ibmvnic_open(struct net_device *netdev)
1443 {
1444 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1445 	int rc;
1446 
1447 	ASSERT_RTNL();
1448 
1449 	/* If device failover is pending or we are about to reset, just set
1450 	 * device state and return. Device operation will be handled by reset
1451 	 * routine.
1452 	 *
1453 	 * It should be safe to overwrite the adapter->state here. Since
1454 	 * we hold the rtnl, either the reset has not actually started or
1455 	 * the rtnl got dropped during the set_link_state() in do_reset().
1456 	 * In the former case, no one else is changing the state (again we
1457 	 * have the rtnl) and in the latter case, do_reset() will detect and
1458 	 * honor our setting below.
1459 	 */
1460 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1461 		netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1462 			   adapter_state_to_string(adapter->state),
1463 			   adapter->failover_pending);
1464 		adapter->state = VNIC_OPEN;
1465 		rc = 0;
1466 		goto out;
1467 	}
1468 
1469 	if (adapter->state != VNIC_CLOSED) {
1470 		rc = ibmvnic_login(netdev);
1471 		if (rc)
1472 			goto out;
1473 
1474 		rc = init_resources(adapter);
1475 		if (rc) {
1476 			netdev_err(netdev, "failed to initialize resources\n");
1477 			release_resources(adapter);
1478 			release_rx_pools(adapter);
1479 			release_tx_pools(adapter);
1480 			goto out;
1481 		}
1482 	}
1483 
1484 	rc = __ibmvnic_open(netdev);
1485 
1486 out:
1487 	/* If open failed and there is a pending failover or in-progress reset,
1488 	 * set device state and return. Device operation will be handled by
1489 	 * reset routine. See also comments above regarding rtnl.
1490 	 */
1491 	if (rc &&
1492 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1493 		adapter->state = VNIC_OPEN;
1494 		rc = 0;
1495 	}
1496 	return rc;
1497 }
1498 
1499 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1500 {
1501 	struct ibmvnic_rx_pool *rx_pool;
1502 	struct ibmvnic_rx_buff *rx_buff;
1503 	u64 rx_entries;
1504 	int rx_scrqs;
1505 	int i, j;
1506 
1507 	if (!adapter->rx_pool)
1508 		return;
1509 
1510 	rx_scrqs = adapter->num_active_rx_pools;
1511 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
1512 
1513 	/* Free any remaining skbs in the rx buffer pools */
1514 	for (i = 0; i < rx_scrqs; i++) {
1515 		rx_pool = &adapter->rx_pool[i];
1516 		if (!rx_pool || !rx_pool->rx_buff)
1517 			continue;
1518 
1519 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1520 		for (j = 0; j < rx_entries; j++) {
1521 			rx_buff = &rx_pool->rx_buff[j];
1522 			if (rx_buff && rx_buff->skb) {
1523 				dev_kfree_skb_any(rx_buff->skb);
1524 				rx_buff->skb = NULL;
1525 			}
1526 		}
1527 	}
1528 }
1529 
1530 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1531 			      struct ibmvnic_tx_pool *tx_pool)
1532 {
1533 	struct ibmvnic_tx_buff *tx_buff;
1534 	u64 tx_entries;
1535 	int i;
1536 
1537 	if (!tx_pool || !tx_pool->tx_buff)
1538 		return;
1539 
1540 	tx_entries = tx_pool->num_buffers;
1541 
1542 	for (i = 0; i < tx_entries; i++) {
1543 		tx_buff = &tx_pool->tx_buff[i];
1544 		if (tx_buff && tx_buff->skb) {
1545 			dev_kfree_skb_any(tx_buff->skb);
1546 			tx_buff->skb = NULL;
1547 		}
1548 	}
1549 }
1550 
1551 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1552 {
1553 	int tx_scrqs;
1554 	int i;
1555 
1556 	if (!adapter->tx_pool || !adapter->tso_pool)
1557 		return;
1558 
1559 	tx_scrqs = adapter->num_active_tx_pools;
1560 
1561 	/* Free any remaining skbs in the tx buffer pools */
1562 	for (i = 0; i < tx_scrqs; i++) {
1563 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1564 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1565 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1566 	}
1567 }
1568 
1569 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1570 {
1571 	struct net_device *netdev = adapter->netdev;
1572 	int i;
1573 
1574 	if (adapter->tx_scrq) {
1575 		for (i = 0; i < adapter->req_tx_queues; i++)
1576 			if (adapter->tx_scrq[i]->irq) {
1577 				netdev_dbg(netdev,
1578 					   "Disabling tx_scrq[%d] irq\n", i);
1579 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1580 				disable_irq(adapter->tx_scrq[i]->irq);
1581 			}
1582 	}
1583 
1584 	if (adapter->rx_scrq) {
1585 		for (i = 0; i < adapter->req_rx_queues; i++) {
1586 			if (adapter->rx_scrq[i]->irq) {
1587 				netdev_dbg(netdev,
1588 					   "Disabling rx_scrq[%d] irq\n", i);
1589 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1590 				disable_irq(adapter->rx_scrq[i]->irq);
1591 			}
1592 		}
1593 	}
1594 }
1595 
1596 static void ibmvnic_cleanup(struct net_device *netdev)
1597 {
1598 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1599 
1600 	/* ensure that transmissions are stopped if called by do_reset */
1601 	if (test_bit(0, &adapter->resetting))
1602 		netif_tx_disable(netdev);
1603 	else
1604 		netif_tx_stop_all_queues(netdev);
1605 
1606 	ibmvnic_napi_disable(adapter);
1607 	ibmvnic_disable_irqs(adapter);
1608 }
1609 
1610 static int __ibmvnic_close(struct net_device *netdev)
1611 {
1612 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1613 	int rc = 0;
1614 
1615 	adapter->state = VNIC_CLOSING;
1616 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1617 	adapter->state = VNIC_CLOSED;
1618 	return rc;
1619 }
1620 
1621 static int ibmvnic_close(struct net_device *netdev)
1622 {
1623 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1624 	int rc;
1625 
1626 	netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1627 		   adapter_state_to_string(adapter->state),
1628 		   adapter->failover_pending,
1629 		   adapter->force_reset_recovery);
1630 
1631 	/* If device failover is pending, just set device state and return.
1632 	 * Device operation will be handled by reset routine.
1633 	 */
1634 	if (adapter->failover_pending) {
1635 		adapter->state = VNIC_CLOSED;
1636 		return 0;
1637 	}
1638 
1639 	rc = __ibmvnic_close(netdev);
1640 	ibmvnic_cleanup(netdev);
1641 	clean_rx_pools(adapter);
1642 	clean_tx_pools(adapter);
1643 
1644 	return rc;
1645 }
1646 
1647 /**
1648  * build_hdr_data - creates L2/L3/L4 header data buffer
1649  * @hdr_field: bitfield determining needed headers
1650  * @skb: socket buffer
1651  * @hdr_len: array of header lengths
1652  * @hdr_data: buffer to write the header to
1653  *
1654  * Reads hdr_field to determine which headers are needed by firmware.
1655  * Builds a buffer containing these headers.  Saves individual header
1656  * lengths and total buffer length to be used to build descriptors.
1657  */
1658 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1659 			  int *hdr_len, u8 *hdr_data)
1660 {
1661 	int len = 0;
1662 	u8 *hdr;
1663 
1664 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1665 		hdr_len[0] = sizeof(struct vlan_ethhdr);
1666 	else
1667 		hdr_len[0] = sizeof(struct ethhdr);
1668 
1669 	if (skb->protocol == htons(ETH_P_IP)) {
1670 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
1671 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1672 			hdr_len[2] = tcp_hdrlen(skb);
1673 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1674 			hdr_len[2] = sizeof(struct udphdr);
1675 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1676 		hdr_len[1] = sizeof(struct ipv6hdr);
1677 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1678 			hdr_len[2] = tcp_hdrlen(skb);
1679 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1680 			hdr_len[2] = sizeof(struct udphdr);
1681 	} else if (skb->protocol == htons(ETH_P_ARP)) {
1682 		hdr_len[1] = arp_hdr_len(skb->dev);
1683 		hdr_len[2] = 0;
1684 	}
1685 
1686 	memset(hdr_data, 0, 120);
1687 	if ((hdr_field >> 6) & 1) {
1688 		hdr = skb_mac_header(skb);
1689 		memcpy(hdr_data, hdr, hdr_len[0]);
1690 		len += hdr_len[0];
1691 	}
1692 
1693 	if ((hdr_field >> 5) & 1) {
1694 		hdr = skb_network_header(skb);
1695 		memcpy(hdr_data + len, hdr, hdr_len[1]);
1696 		len += hdr_len[1];
1697 	}
1698 
1699 	if ((hdr_field >> 4) & 1) {
1700 		hdr = skb_transport_header(skb);
1701 		memcpy(hdr_data + len, hdr, hdr_len[2]);
1702 		len += hdr_len[2];
1703 	}
1704 	return len;
1705 }
1706 
1707 /**
1708  * create_hdr_descs - create header and header extension descriptors
1709  * @hdr_field: bitfield determining needed headers
1710  * @hdr_data: buffer containing header data
1711  * @len: length of data buffer
1712  * @hdr_len: array of individual header lengths
1713  * @scrq_arr: descriptor array
1714  *
1715  * Creates header and, if needed, header extension descriptors and
1716  * places them in a descriptor array, scrq_arr
1717  */
1718 
1719 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1720 			    union sub_crq *scrq_arr)
1721 {
1722 	union sub_crq hdr_desc;
1723 	int tmp_len = len;
1724 	int num_descs = 0;
1725 	u8 *data, *cur;
1726 	int tmp;
1727 
1728 	while (tmp_len > 0) {
1729 		cur = hdr_data + len - tmp_len;
1730 
1731 		memset(&hdr_desc, 0, sizeof(hdr_desc));
1732 		if (cur != hdr_data) {
1733 			data = hdr_desc.hdr_ext.data;
1734 			tmp = tmp_len > 29 ? 29 : tmp_len;
1735 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1736 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1737 			hdr_desc.hdr_ext.len = tmp;
1738 		} else {
1739 			data = hdr_desc.hdr.data;
1740 			tmp = tmp_len > 24 ? 24 : tmp_len;
1741 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1742 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1743 			hdr_desc.hdr.len = tmp;
1744 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1745 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1746 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1747 			hdr_desc.hdr.flag = hdr_field << 1;
1748 		}
1749 		memcpy(data, cur, tmp);
1750 		tmp_len -= tmp;
1751 		*scrq_arr = hdr_desc;
1752 		scrq_arr++;
1753 		num_descs++;
1754 	}
1755 
1756 	return num_descs;
1757 }
1758 
1759 /**
1760  * build_hdr_descs_arr - build a header descriptor array
1761  * @skb: tx socket buffer
1762  * @indir_arr: indirect array
1763  * @num_entries: number of descriptors to be sent
1764  * @hdr_field: bit field determining which headers will be sent
1765  *
1766  * This function will build a TX descriptor array with applicable
1767  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1768  */
1769 
1770 static void build_hdr_descs_arr(struct sk_buff *skb,
1771 				union sub_crq *indir_arr,
1772 				int *num_entries, u8 hdr_field)
1773 {
1774 	int hdr_len[3] = {0, 0, 0};
1775 	u8 hdr_data[140] = {0};
1776 	int tot_len;
1777 
1778 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1779 				 hdr_data);
1780 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1781 					 indir_arr + 1);
1782 }
1783 
1784 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1785 				    struct net_device *netdev)
1786 {
1787 	/* For some backing devices, mishandling of small packets
1788 	 * can result in a loss of connection or TX stall. Device
1789 	 * architects recommend that no packet should be smaller
1790 	 * than the minimum MTU value provided to the driver, so
1791 	 * pad any packets to that length
1792 	 */
1793 	if (skb->len < netdev->min_mtu)
1794 		return skb_put_padto(skb, netdev->min_mtu);
1795 
1796 	return 0;
1797 }
1798 
1799 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1800 					 struct ibmvnic_sub_crq_queue *tx_scrq)
1801 {
1802 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1803 	struct ibmvnic_tx_buff *tx_buff;
1804 	struct ibmvnic_tx_pool *tx_pool;
1805 	union sub_crq tx_scrq_entry;
1806 	int queue_num;
1807 	int entries;
1808 	int index;
1809 	int i;
1810 
1811 	ind_bufp = &tx_scrq->ind_buf;
1812 	entries = (u64)ind_bufp->index;
1813 	queue_num = tx_scrq->pool_index;
1814 
1815 	for (i = entries - 1; i >= 0; --i) {
1816 		tx_scrq_entry = ind_bufp->indir_arr[i];
1817 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1818 			continue;
1819 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1820 		if (index & IBMVNIC_TSO_POOL_MASK) {
1821 			tx_pool = &adapter->tso_pool[queue_num];
1822 			index &= ~IBMVNIC_TSO_POOL_MASK;
1823 		} else {
1824 			tx_pool = &adapter->tx_pool[queue_num];
1825 		}
1826 		tx_pool->free_map[tx_pool->consumer_index] = index;
1827 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1828 					  tx_pool->num_buffers - 1 :
1829 					  tx_pool->consumer_index - 1;
1830 		tx_buff = &tx_pool->tx_buff[index];
1831 		adapter->netdev->stats.tx_packets--;
1832 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1833 		adapter->tx_stats_buffers[queue_num].packets--;
1834 		adapter->tx_stats_buffers[queue_num].bytes -=
1835 						tx_buff->skb->len;
1836 		dev_kfree_skb_any(tx_buff->skb);
1837 		tx_buff->skb = NULL;
1838 		adapter->netdev->stats.tx_dropped++;
1839 	}
1840 	ind_bufp->index = 0;
1841 	if (atomic_sub_return(entries, &tx_scrq->used) <=
1842 	    (adapter->req_tx_entries_per_subcrq / 2) &&
1843 	    __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1844 	    !test_bit(0, &adapter->resetting)) {
1845 		netif_wake_subqueue(adapter->netdev, queue_num);
1846 		netdev_dbg(adapter->netdev, "Started queue %d\n",
1847 			   queue_num);
1848 	}
1849 }
1850 
1851 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1852 				 struct ibmvnic_sub_crq_queue *tx_scrq)
1853 {
1854 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1855 	u64 dma_addr;
1856 	u64 entries;
1857 	u64 handle;
1858 	int rc;
1859 
1860 	ind_bufp = &tx_scrq->ind_buf;
1861 	dma_addr = (u64)ind_bufp->indir_dma;
1862 	entries = (u64)ind_bufp->index;
1863 	handle = tx_scrq->handle;
1864 
1865 	if (!entries)
1866 		return 0;
1867 	rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1868 	if (rc)
1869 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1870 	else
1871 		ind_bufp->index = 0;
1872 	return 0;
1873 }
1874 
1875 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1876 {
1877 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1878 	int queue_num = skb_get_queue_mapping(skb);
1879 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1880 	struct device *dev = &adapter->vdev->dev;
1881 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1882 	struct ibmvnic_tx_buff *tx_buff = NULL;
1883 	struct ibmvnic_sub_crq_queue *tx_scrq;
1884 	struct ibmvnic_tx_pool *tx_pool;
1885 	unsigned int tx_send_failed = 0;
1886 	netdev_tx_t ret = NETDEV_TX_OK;
1887 	unsigned int tx_map_failed = 0;
1888 	union sub_crq indir_arr[16];
1889 	unsigned int tx_dropped = 0;
1890 	unsigned int tx_packets = 0;
1891 	unsigned int tx_bytes = 0;
1892 	dma_addr_t data_dma_addr;
1893 	struct netdev_queue *txq;
1894 	unsigned long lpar_rc;
1895 	union sub_crq tx_crq;
1896 	unsigned int offset;
1897 	int num_entries = 1;
1898 	unsigned char *dst;
1899 	int index = 0;
1900 	u8 proto = 0;
1901 
1902 	tx_scrq = adapter->tx_scrq[queue_num];
1903 	txq = netdev_get_tx_queue(netdev, queue_num);
1904 	ind_bufp = &tx_scrq->ind_buf;
1905 
1906 	if (test_bit(0, &adapter->resetting)) {
1907 		dev_kfree_skb_any(skb);
1908 
1909 		tx_send_failed++;
1910 		tx_dropped++;
1911 		ret = NETDEV_TX_OK;
1912 		goto out;
1913 	}
1914 
1915 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
1916 		tx_dropped++;
1917 		tx_send_failed++;
1918 		ret = NETDEV_TX_OK;
1919 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1920 		goto out;
1921 	}
1922 	if (skb_is_gso(skb))
1923 		tx_pool = &adapter->tso_pool[queue_num];
1924 	else
1925 		tx_pool = &adapter->tx_pool[queue_num];
1926 
1927 	index = tx_pool->free_map[tx_pool->consumer_index];
1928 
1929 	if (index == IBMVNIC_INVALID_MAP) {
1930 		dev_kfree_skb_any(skb);
1931 		tx_send_failed++;
1932 		tx_dropped++;
1933 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1934 		ret = NETDEV_TX_OK;
1935 		goto out;
1936 	}
1937 
1938 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1939 
1940 	offset = index * tx_pool->buf_size;
1941 	dst = tx_pool->long_term_buff.buff + offset;
1942 	memset(dst, 0, tx_pool->buf_size);
1943 	data_dma_addr = tx_pool->long_term_buff.addr + offset;
1944 
1945 	if (skb_shinfo(skb)->nr_frags) {
1946 		int cur, i;
1947 
1948 		/* Copy the head */
1949 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1950 		cur = skb_headlen(skb);
1951 
1952 		/* Copy the frags */
1953 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1954 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1955 
1956 			memcpy(dst + cur, skb_frag_address(frag),
1957 			       skb_frag_size(frag));
1958 			cur += skb_frag_size(frag);
1959 		}
1960 	} else {
1961 		skb_copy_from_linear_data(skb, dst, skb->len);
1962 	}
1963 
1964 	/* post changes to long_term_buff *dst before VIOS accessing it */
1965 	dma_wmb();
1966 
1967 	tx_pool->consumer_index =
1968 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1969 
1970 	tx_buff = &tx_pool->tx_buff[index];
1971 	tx_buff->skb = skb;
1972 	tx_buff->index = index;
1973 	tx_buff->pool_index = queue_num;
1974 
1975 	memset(&tx_crq, 0, sizeof(tx_crq));
1976 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1977 	tx_crq.v1.type = IBMVNIC_TX_DESC;
1978 	tx_crq.v1.n_crq_elem = 1;
1979 	tx_crq.v1.n_sge = 1;
1980 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1981 
1982 	if (skb_is_gso(skb))
1983 		tx_crq.v1.correlator =
1984 			cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1985 	else
1986 		tx_crq.v1.correlator = cpu_to_be32(index);
1987 	tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1988 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1989 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1990 
1991 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1992 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1993 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1994 	}
1995 
1996 	if (skb->protocol == htons(ETH_P_IP)) {
1997 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1998 		proto = ip_hdr(skb)->protocol;
1999 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2000 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2001 		proto = ipv6_hdr(skb)->nexthdr;
2002 	}
2003 
2004 	if (proto == IPPROTO_TCP)
2005 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2006 	else if (proto == IPPROTO_UDP)
2007 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2008 
2009 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2010 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2011 		hdrs += 2;
2012 	}
2013 	if (skb_is_gso(skb)) {
2014 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2015 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2016 		hdrs += 2;
2017 	}
2018 
2019 	if ((*hdrs >> 7) & 1)
2020 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2021 
2022 	tx_crq.v1.n_crq_elem = num_entries;
2023 	tx_buff->num_entries = num_entries;
2024 	/* flush buffer if current entry can not fit */
2025 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2026 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2027 		if (lpar_rc != H_SUCCESS)
2028 			goto tx_flush_err;
2029 	}
2030 
2031 	indir_arr[0] = tx_crq;
2032 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2033 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2034 	ind_bufp->index += num_entries;
2035 	if (__netdev_tx_sent_queue(txq, skb->len,
2036 				   netdev_xmit_more() &&
2037 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2038 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2039 		if (lpar_rc != H_SUCCESS)
2040 			goto tx_err;
2041 	}
2042 
2043 	if (atomic_add_return(num_entries, &tx_scrq->used)
2044 					>= adapter->req_tx_entries_per_subcrq) {
2045 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2046 		netif_stop_subqueue(netdev, queue_num);
2047 	}
2048 
2049 	tx_packets++;
2050 	tx_bytes += skb->len;
2051 	txq_trans_cond_update(txq);
2052 	ret = NETDEV_TX_OK;
2053 	goto out;
2054 
2055 tx_flush_err:
2056 	dev_kfree_skb_any(skb);
2057 	tx_buff->skb = NULL;
2058 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2059 				  tx_pool->num_buffers - 1 :
2060 				  tx_pool->consumer_index - 1;
2061 	tx_dropped++;
2062 tx_err:
2063 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2064 		dev_err_ratelimited(dev, "tx: send failed\n");
2065 
2066 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2067 		/* Disable TX and report carrier off if queue is closed
2068 		 * or pending failover.
2069 		 * Firmware guarantees that a signal will be sent to the
2070 		 * driver, triggering a reset or some other action.
2071 		 */
2072 		netif_tx_stop_all_queues(netdev);
2073 		netif_carrier_off(netdev);
2074 	}
2075 out:
2076 	netdev->stats.tx_dropped += tx_dropped;
2077 	netdev->stats.tx_bytes += tx_bytes;
2078 	netdev->stats.tx_packets += tx_packets;
2079 	adapter->tx_send_failed += tx_send_failed;
2080 	adapter->tx_map_failed += tx_map_failed;
2081 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2082 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2083 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2084 
2085 	return ret;
2086 }
2087 
2088 static void ibmvnic_set_multi(struct net_device *netdev)
2089 {
2090 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2091 	struct netdev_hw_addr *ha;
2092 	union ibmvnic_crq crq;
2093 
2094 	memset(&crq, 0, sizeof(crq));
2095 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2096 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2097 
2098 	if (netdev->flags & IFF_PROMISC) {
2099 		if (!adapter->promisc_supported)
2100 			return;
2101 	} else {
2102 		if (netdev->flags & IFF_ALLMULTI) {
2103 			/* Accept all multicast */
2104 			memset(&crq, 0, sizeof(crq));
2105 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2106 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2107 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2108 			ibmvnic_send_crq(adapter, &crq);
2109 		} else if (netdev_mc_empty(netdev)) {
2110 			/* Reject all multicast */
2111 			memset(&crq, 0, sizeof(crq));
2112 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2113 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2114 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2115 			ibmvnic_send_crq(adapter, &crq);
2116 		} else {
2117 			/* Accept one or more multicast(s) */
2118 			netdev_for_each_mc_addr(ha, netdev) {
2119 				memset(&crq, 0, sizeof(crq));
2120 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2121 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2122 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2123 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2124 						ha->addr);
2125 				ibmvnic_send_crq(adapter, &crq);
2126 			}
2127 		}
2128 	}
2129 }
2130 
2131 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2132 {
2133 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2134 	union ibmvnic_crq crq;
2135 	int rc;
2136 
2137 	if (!is_valid_ether_addr(dev_addr)) {
2138 		rc = -EADDRNOTAVAIL;
2139 		goto err;
2140 	}
2141 
2142 	memset(&crq, 0, sizeof(crq));
2143 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2144 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2145 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2146 
2147 	mutex_lock(&adapter->fw_lock);
2148 	adapter->fw_done_rc = 0;
2149 	reinit_completion(&adapter->fw_done);
2150 
2151 	rc = ibmvnic_send_crq(adapter, &crq);
2152 	if (rc) {
2153 		rc = -EIO;
2154 		mutex_unlock(&adapter->fw_lock);
2155 		goto err;
2156 	}
2157 
2158 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2159 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2160 	if (rc || adapter->fw_done_rc) {
2161 		rc = -EIO;
2162 		mutex_unlock(&adapter->fw_lock);
2163 		goto err;
2164 	}
2165 	mutex_unlock(&adapter->fw_lock);
2166 	return 0;
2167 err:
2168 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2169 	return rc;
2170 }
2171 
2172 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2173 {
2174 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2175 	struct sockaddr *addr = p;
2176 	int rc;
2177 
2178 	rc = 0;
2179 	if (!is_valid_ether_addr(addr->sa_data))
2180 		return -EADDRNOTAVAIL;
2181 
2182 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2183 	if (adapter->state != VNIC_PROBED)
2184 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2185 
2186 	return rc;
2187 }
2188 
2189 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2190 {
2191 	switch (reason) {
2192 	case VNIC_RESET_FAILOVER:
2193 		return "FAILOVER";
2194 	case VNIC_RESET_MOBILITY:
2195 		return "MOBILITY";
2196 	case VNIC_RESET_FATAL:
2197 		return "FATAL";
2198 	case VNIC_RESET_NON_FATAL:
2199 		return "NON_FATAL";
2200 	case VNIC_RESET_TIMEOUT:
2201 		return "TIMEOUT";
2202 	case VNIC_RESET_CHANGE_PARAM:
2203 		return "CHANGE_PARAM";
2204 	case VNIC_RESET_PASSIVE_INIT:
2205 		return "PASSIVE_INIT";
2206 	}
2207 	return "UNKNOWN";
2208 }
2209 
2210 /*
2211  * do_reset returns zero if we are able to keep processing reset events, or
2212  * non-zero if we hit a fatal error and must halt.
2213  */
2214 static int do_reset(struct ibmvnic_adapter *adapter,
2215 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2216 {
2217 	struct net_device *netdev = adapter->netdev;
2218 	u64 old_num_rx_queues, old_num_tx_queues;
2219 	u64 old_num_rx_slots, old_num_tx_slots;
2220 	int rc;
2221 
2222 	netdev_dbg(adapter->netdev,
2223 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2224 		   adapter_state_to_string(adapter->state),
2225 		   adapter->failover_pending,
2226 		   reset_reason_to_string(rwi->reset_reason),
2227 		   adapter_state_to_string(reset_state));
2228 
2229 	adapter->reset_reason = rwi->reset_reason;
2230 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2231 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2232 		rtnl_lock();
2233 
2234 	/* Now that we have the rtnl lock, clear any pending failover.
2235 	 * This will ensure ibmvnic_open() has either completed or will
2236 	 * block until failover is complete.
2237 	 */
2238 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2239 		adapter->failover_pending = false;
2240 
2241 	/* read the state and check (again) after getting rtnl */
2242 	reset_state = adapter->state;
2243 
2244 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2245 		rc = -EBUSY;
2246 		goto out;
2247 	}
2248 
2249 	netif_carrier_off(netdev);
2250 
2251 	old_num_rx_queues = adapter->req_rx_queues;
2252 	old_num_tx_queues = adapter->req_tx_queues;
2253 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2254 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2255 
2256 	ibmvnic_cleanup(netdev);
2257 
2258 	if (reset_state == VNIC_OPEN &&
2259 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2260 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2261 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2262 			rc = __ibmvnic_close(netdev);
2263 			if (rc)
2264 				goto out;
2265 		} else {
2266 			adapter->state = VNIC_CLOSING;
2267 
2268 			/* Release the RTNL lock before link state change and
2269 			 * re-acquire after the link state change to allow
2270 			 * linkwatch_event to grab the RTNL lock and run during
2271 			 * a reset.
2272 			 */
2273 			rtnl_unlock();
2274 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2275 			rtnl_lock();
2276 			if (rc)
2277 				goto out;
2278 
2279 			if (adapter->state == VNIC_OPEN) {
2280 				/* When we dropped rtnl, ibmvnic_open() got
2281 				 * it and noticed that we are resetting and
2282 				 * set the adapter state to OPEN. Update our
2283 				 * new "target" state, and resume the reset
2284 				 * from VNIC_CLOSING state.
2285 				 */
2286 				netdev_dbg(netdev,
2287 					   "Open changed state from %s, updating.\n",
2288 					   adapter_state_to_string(reset_state));
2289 				reset_state = VNIC_OPEN;
2290 				adapter->state = VNIC_CLOSING;
2291 			}
2292 
2293 			if (adapter->state != VNIC_CLOSING) {
2294 				/* If someone else changed the adapter state
2295 				 * when we dropped the rtnl, fail the reset
2296 				 */
2297 				rc = -EAGAIN;
2298 				goto out;
2299 			}
2300 			adapter->state = VNIC_CLOSED;
2301 		}
2302 	}
2303 
2304 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2305 		release_resources(adapter);
2306 		release_sub_crqs(adapter, 1);
2307 		release_crq_queue(adapter);
2308 	}
2309 
2310 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2311 		/* remove the closed state so when we call open it appears
2312 		 * we are coming from the probed state.
2313 		 */
2314 		adapter->state = VNIC_PROBED;
2315 
2316 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2317 			rc = init_crq_queue(adapter);
2318 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2319 			rc = ibmvnic_reenable_crq_queue(adapter);
2320 			release_sub_crqs(adapter, 1);
2321 		} else {
2322 			rc = ibmvnic_reset_crq(adapter);
2323 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2324 				rc = vio_enable_interrupts(adapter->vdev);
2325 				if (rc)
2326 					netdev_err(adapter->netdev,
2327 						   "Reset failed to enable interrupts. rc=%d\n",
2328 						   rc);
2329 			}
2330 		}
2331 
2332 		if (rc) {
2333 			netdev_err(adapter->netdev,
2334 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2335 			goto out;
2336 		}
2337 
2338 		rc = ibmvnic_reset_init(adapter, true);
2339 		if (rc)
2340 			goto out;
2341 
2342 		/* If the adapter was in PROBE or DOWN state prior to the reset,
2343 		 * exit here.
2344 		 */
2345 		if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2346 			rc = 0;
2347 			goto out;
2348 		}
2349 
2350 		rc = ibmvnic_login(netdev);
2351 		if (rc)
2352 			goto out;
2353 
2354 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2355 			rc = init_resources(adapter);
2356 			if (rc)
2357 				goto out;
2358 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2359 		    adapter->req_tx_queues != old_num_tx_queues ||
2360 		    adapter->req_rx_add_entries_per_subcrq !=
2361 		    old_num_rx_slots ||
2362 		    adapter->req_tx_entries_per_subcrq !=
2363 		    old_num_tx_slots ||
2364 		    !adapter->rx_pool ||
2365 		    !adapter->tso_pool ||
2366 		    !adapter->tx_pool) {
2367 			release_napi(adapter);
2368 			release_vpd_data(adapter);
2369 
2370 			rc = init_resources(adapter);
2371 			if (rc)
2372 				goto out;
2373 
2374 		} else {
2375 			rc = init_tx_pools(netdev);
2376 			if (rc) {
2377 				netdev_dbg(netdev,
2378 					   "init tx pools failed (%d)\n",
2379 					   rc);
2380 				goto out;
2381 			}
2382 
2383 			rc = init_rx_pools(netdev);
2384 			if (rc) {
2385 				netdev_dbg(netdev,
2386 					   "init rx pools failed (%d)\n",
2387 					   rc);
2388 				goto out;
2389 			}
2390 		}
2391 		ibmvnic_disable_irqs(adapter);
2392 	}
2393 	adapter->state = VNIC_CLOSED;
2394 
2395 	if (reset_state == VNIC_CLOSED) {
2396 		rc = 0;
2397 		goto out;
2398 	}
2399 
2400 	rc = __ibmvnic_open(netdev);
2401 	if (rc) {
2402 		rc = IBMVNIC_OPEN_FAILED;
2403 		goto out;
2404 	}
2405 
2406 	/* refresh device's multicast list */
2407 	ibmvnic_set_multi(netdev);
2408 
2409 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2410 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2411 		__netdev_notify_peers(netdev);
2412 
2413 	rc = 0;
2414 
2415 out:
2416 	/* restore the adapter state if reset failed */
2417 	if (rc)
2418 		adapter->state = reset_state;
2419 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2420 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2421 		rtnl_unlock();
2422 
2423 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2424 		   adapter_state_to_string(adapter->state),
2425 		   adapter->failover_pending, rc);
2426 	return rc;
2427 }
2428 
2429 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2430 			 struct ibmvnic_rwi *rwi, u32 reset_state)
2431 {
2432 	struct net_device *netdev = adapter->netdev;
2433 	int rc;
2434 
2435 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2436 		   reset_reason_to_string(rwi->reset_reason));
2437 
2438 	/* read the state and check (again) after getting rtnl */
2439 	reset_state = adapter->state;
2440 
2441 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2442 		rc = -EBUSY;
2443 		goto out;
2444 	}
2445 
2446 	netif_carrier_off(netdev);
2447 	adapter->reset_reason = rwi->reset_reason;
2448 
2449 	ibmvnic_cleanup(netdev);
2450 	release_resources(adapter);
2451 	release_sub_crqs(adapter, 0);
2452 	release_crq_queue(adapter);
2453 
2454 	/* remove the closed state so when we call open it appears
2455 	 * we are coming from the probed state.
2456 	 */
2457 	adapter->state = VNIC_PROBED;
2458 
2459 	reinit_completion(&adapter->init_done);
2460 	rc = init_crq_queue(adapter);
2461 	if (rc) {
2462 		netdev_err(adapter->netdev,
2463 			   "Couldn't initialize crq. rc=%d\n", rc);
2464 		goto out;
2465 	}
2466 
2467 	rc = ibmvnic_reset_init(adapter, false);
2468 	if (rc)
2469 		goto out;
2470 
2471 	/* If the adapter was in PROBE or DOWN state prior to the reset,
2472 	 * exit here.
2473 	 */
2474 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2475 		goto out;
2476 
2477 	rc = ibmvnic_login(netdev);
2478 	if (rc)
2479 		goto out;
2480 
2481 	rc = init_resources(adapter);
2482 	if (rc)
2483 		goto out;
2484 
2485 	ibmvnic_disable_irqs(adapter);
2486 	adapter->state = VNIC_CLOSED;
2487 
2488 	if (reset_state == VNIC_CLOSED)
2489 		goto out;
2490 
2491 	rc = __ibmvnic_open(netdev);
2492 	if (rc) {
2493 		rc = IBMVNIC_OPEN_FAILED;
2494 		goto out;
2495 	}
2496 
2497 	__netdev_notify_peers(netdev);
2498 out:
2499 	/* restore adapter state if reset failed */
2500 	if (rc)
2501 		adapter->state = reset_state;
2502 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2503 		   adapter_state_to_string(adapter->state),
2504 		   adapter->failover_pending, rc);
2505 	return rc;
2506 }
2507 
2508 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2509 {
2510 	struct ibmvnic_rwi *rwi;
2511 	unsigned long flags;
2512 
2513 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2514 
2515 	if (!list_empty(&adapter->rwi_list)) {
2516 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2517 				       list);
2518 		list_del(&rwi->list);
2519 	} else {
2520 		rwi = NULL;
2521 	}
2522 
2523 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2524 	return rwi;
2525 }
2526 
2527 /**
2528  * do_passive_init - complete probing when partner device is detected.
2529  * @adapter: ibmvnic_adapter struct
2530  *
2531  * If the ibmvnic device does not have a partner device to communicate with at boot
2532  * and that partner device comes online at a later time, this function is called
2533  * to complete the initialization process of ibmvnic device.
2534  * Caller is expected to hold rtnl_lock().
2535  *
2536  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2537  * in the down state.
2538  * Returns 0 upon success and the device is in PROBED state.
2539  */
2540 
2541 static int do_passive_init(struct ibmvnic_adapter *adapter)
2542 {
2543 	unsigned long timeout = msecs_to_jiffies(30000);
2544 	struct net_device *netdev = adapter->netdev;
2545 	struct device *dev = &adapter->vdev->dev;
2546 	int rc;
2547 
2548 	netdev_dbg(netdev, "Partner device found, probing.\n");
2549 
2550 	adapter->state = VNIC_PROBING;
2551 	reinit_completion(&adapter->init_done);
2552 	adapter->init_done_rc = 0;
2553 	adapter->crq.active = true;
2554 
2555 	rc = send_crq_init_complete(adapter);
2556 	if (rc)
2557 		goto out;
2558 
2559 	rc = send_version_xchg(adapter);
2560 	if (rc)
2561 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2562 
2563 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2564 		dev_err(dev, "Initialization sequence timed out\n");
2565 		rc = -ETIMEDOUT;
2566 		goto out;
2567 	}
2568 
2569 	rc = init_sub_crqs(adapter);
2570 	if (rc) {
2571 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2572 		goto out;
2573 	}
2574 
2575 	rc = init_sub_crq_irqs(adapter);
2576 	if (rc) {
2577 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2578 		goto init_failed;
2579 	}
2580 
2581 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
2582 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2583 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2584 
2585 	adapter->state = VNIC_PROBED;
2586 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2587 
2588 	return 0;
2589 
2590 init_failed:
2591 	release_sub_crqs(adapter, 1);
2592 out:
2593 	adapter->state = VNIC_DOWN;
2594 	return rc;
2595 }
2596 
2597 static void __ibmvnic_reset(struct work_struct *work)
2598 {
2599 	struct ibmvnic_adapter *adapter;
2600 	bool saved_state = false;
2601 	struct ibmvnic_rwi *tmprwi;
2602 	struct ibmvnic_rwi *rwi;
2603 	unsigned long flags;
2604 	u32 reset_state;
2605 	int num_fails = 0;
2606 	int rc = 0;
2607 
2608 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2609 
2610 	if (test_and_set_bit_lock(0, &adapter->resetting)) {
2611 		queue_delayed_work(system_long_wq,
2612 				   &adapter->ibmvnic_delayed_reset,
2613 				   IBMVNIC_RESET_DELAY);
2614 		return;
2615 	}
2616 
2617 	rwi = get_next_rwi(adapter);
2618 	while (rwi) {
2619 		spin_lock_irqsave(&adapter->state_lock, flags);
2620 
2621 		if (adapter->state == VNIC_REMOVING ||
2622 		    adapter->state == VNIC_REMOVED) {
2623 			spin_unlock_irqrestore(&adapter->state_lock, flags);
2624 			kfree(rwi);
2625 			rc = EBUSY;
2626 			break;
2627 		}
2628 
2629 		if (!saved_state) {
2630 			reset_state = adapter->state;
2631 			saved_state = true;
2632 		}
2633 		spin_unlock_irqrestore(&adapter->state_lock, flags);
2634 
2635 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2636 			rtnl_lock();
2637 			rc = do_passive_init(adapter);
2638 			rtnl_unlock();
2639 			if (!rc)
2640 				netif_carrier_on(adapter->netdev);
2641 		} else if (adapter->force_reset_recovery) {
2642 			/* Since we are doing a hard reset now, clear the
2643 			 * failover_pending flag so we don't ignore any
2644 			 * future MOBILITY or other resets.
2645 			 */
2646 			adapter->failover_pending = false;
2647 
2648 			/* Transport event occurred during previous reset */
2649 			if (adapter->wait_for_reset) {
2650 				/* Previous was CHANGE_PARAM; caller locked */
2651 				adapter->force_reset_recovery = false;
2652 				rc = do_hard_reset(adapter, rwi, reset_state);
2653 			} else {
2654 				rtnl_lock();
2655 				adapter->force_reset_recovery = false;
2656 				rc = do_hard_reset(adapter, rwi, reset_state);
2657 				rtnl_unlock();
2658 			}
2659 			if (rc)
2660 				num_fails++;
2661 			else
2662 				num_fails = 0;
2663 
2664 			/* If auto-priority-failover is enabled we can get
2665 			 * back to back failovers during resets, resulting
2666 			 * in at least two failed resets (from high-priority
2667 			 * backing device to low-priority one and then back)
2668 			 * If resets continue to fail beyond that, give the
2669 			 * adapter some time to settle down before retrying.
2670 			 */
2671 			if (num_fails >= 3) {
2672 				netdev_dbg(adapter->netdev,
2673 					   "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
2674 					   adapter_state_to_string(adapter->state),
2675 					   num_fails);
2676 				set_current_state(TASK_UNINTERRUPTIBLE);
2677 				schedule_timeout(60 * HZ);
2678 			}
2679 		} else {
2680 			rc = do_reset(adapter, rwi, reset_state);
2681 		}
2682 		tmprwi = rwi;
2683 		adapter->last_reset_time = jiffies;
2684 
2685 		if (rc)
2686 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2687 
2688 		rwi = get_next_rwi(adapter);
2689 
2690 		/*
2691 		 * If there is another reset queued, free the previous rwi
2692 		 * and process the new reset even if previous reset failed
2693 		 * (the previous reset could have failed because of a fail
2694 		 * over for instance, so process the fail over).
2695 		 *
2696 		 * If there are no resets queued and the previous reset failed,
2697 		 * the adapter would be in an undefined state. So retry the
2698 		 * previous reset as a hard reset.
2699 		 */
2700 		if (rwi)
2701 			kfree(tmprwi);
2702 		else if (rc)
2703 			rwi = tmprwi;
2704 
2705 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2706 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2707 			adapter->force_reset_recovery = true;
2708 	}
2709 
2710 	if (adapter->wait_for_reset) {
2711 		adapter->reset_done_rc = rc;
2712 		complete(&adapter->reset_done);
2713 	}
2714 
2715 	clear_bit_unlock(0, &adapter->resetting);
2716 
2717 	netdev_dbg(adapter->netdev,
2718 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2719 		   adapter_state_to_string(adapter->state),
2720 		   adapter->force_reset_recovery,
2721 		   adapter->wait_for_reset);
2722 }
2723 
2724 static void __ibmvnic_delayed_reset(struct work_struct *work)
2725 {
2726 	struct ibmvnic_adapter *adapter;
2727 
2728 	adapter = container_of(work, struct ibmvnic_adapter,
2729 			       ibmvnic_delayed_reset.work);
2730 	__ibmvnic_reset(&adapter->ibmvnic_reset);
2731 }
2732 
2733 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2734 			 enum ibmvnic_reset_reason reason)
2735 {
2736 	struct list_head *entry, *tmp_entry;
2737 	struct ibmvnic_rwi *rwi, *tmp;
2738 	struct net_device *netdev = adapter->netdev;
2739 	unsigned long flags;
2740 	int ret;
2741 
2742 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2743 
2744 	/* If failover is pending don't schedule any other reset.
2745 	 * Instead let the failover complete. If there is already a
2746 	 * a failover reset scheduled, we will detect and drop the
2747 	 * duplicate reset when walking the ->rwi_list below.
2748 	 */
2749 	if (adapter->state == VNIC_REMOVING ||
2750 	    adapter->state == VNIC_REMOVED ||
2751 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2752 		ret = EBUSY;
2753 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2754 		goto err;
2755 	}
2756 
2757 	if (adapter->state == VNIC_PROBING) {
2758 		netdev_warn(netdev, "Adapter reset during probe\n");
2759 		adapter->init_done_rc = -EAGAIN;
2760 		ret = EAGAIN;
2761 		goto err;
2762 	}
2763 
2764 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
2765 		if (tmp->reset_reason == reason) {
2766 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2767 				   reset_reason_to_string(reason));
2768 			ret = EBUSY;
2769 			goto err;
2770 		}
2771 	}
2772 
2773 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2774 	if (!rwi) {
2775 		ret = ENOMEM;
2776 		goto err;
2777 	}
2778 	/* if we just received a transport event,
2779 	 * flush reset queue and process this reset
2780 	 */
2781 	if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2782 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2783 			list_del(entry);
2784 	}
2785 	rwi->reset_reason = reason;
2786 	list_add_tail(&rwi->list, &adapter->rwi_list);
2787 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2788 		   reset_reason_to_string(reason));
2789 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
2790 
2791 	ret = 0;
2792 err:
2793 	/* ibmvnic_close() below can block, so drop the lock first */
2794 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2795 
2796 	if (ret == ENOMEM)
2797 		ibmvnic_close(netdev);
2798 
2799 	return -ret;
2800 }
2801 
2802 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2803 {
2804 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2805 
2806 	if (test_bit(0, &adapter->resetting)) {
2807 		netdev_err(adapter->netdev,
2808 			   "Adapter is resetting, skip timeout reset\n");
2809 		return;
2810 	}
2811 	/* No queuing up reset until at least 5 seconds (default watchdog val)
2812 	 * after last reset
2813 	 */
2814 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2815 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
2816 		return;
2817 	}
2818 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2819 }
2820 
2821 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2822 				  struct ibmvnic_rx_buff *rx_buff)
2823 {
2824 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2825 
2826 	rx_buff->skb = NULL;
2827 
2828 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2829 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2830 
2831 	atomic_dec(&pool->available);
2832 }
2833 
2834 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2835 {
2836 	struct ibmvnic_sub_crq_queue *rx_scrq;
2837 	struct ibmvnic_adapter *adapter;
2838 	struct net_device *netdev;
2839 	int frames_processed;
2840 	int scrq_num;
2841 
2842 	netdev = napi->dev;
2843 	adapter = netdev_priv(netdev);
2844 	scrq_num = (int)(napi - adapter->napi);
2845 	frames_processed = 0;
2846 	rx_scrq = adapter->rx_scrq[scrq_num];
2847 
2848 restart_poll:
2849 	while (frames_processed < budget) {
2850 		struct sk_buff *skb;
2851 		struct ibmvnic_rx_buff *rx_buff;
2852 		union sub_crq *next;
2853 		u32 length;
2854 		u16 offset;
2855 		u8 flags = 0;
2856 
2857 		if (unlikely(test_bit(0, &adapter->resetting) &&
2858 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2859 			enable_scrq_irq(adapter, rx_scrq);
2860 			napi_complete_done(napi, frames_processed);
2861 			return frames_processed;
2862 		}
2863 
2864 		if (!pending_scrq(adapter, rx_scrq))
2865 			break;
2866 		next = ibmvnic_next_scrq(adapter, rx_scrq);
2867 		rx_buff = (struct ibmvnic_rx_buff *)
2868 			  be64_to_cpu(next->rx_comp.correlator);
2869 		/* do error checking */
2870 		if (next->rx_comp.rc) {
2871 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2872 				   be16_to_cpu(next->rx_comp.rc));
2873 			/* free the entry */
2874 			next->rx_comp.first = 0;
2875 			dev_kfree_skb_any(rx_buff->skb);
2876 			remove_buff_from_pool(adapter, rx_buff);
2877 			continue;
2878 		} else if (!rx_buff->skb) {
2879 			/* free the entry */
2880 			next->rx_comp.first = 0;
2881 			remove_buff_from_pool(adapter, rx_buff);
2882 			continue;
2883 		}
2884 
2885 		length = be32_to_cpu(next->rx_comp.len);
2886 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
2887 		flags = next->rx_comp.flags;
2888 		skb = rx_buff->skb;
2889 		/* load long_term_buff before copying to skb */
2890 		dma_rmb();
2891 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
2892 					length);
2893 
2894 		/* VLAN Header has been stripped by the system firmware and
2895 		 * needs to be inserted by the driver
2896 		 */
2897 		if (adapter->rx_vlan_header_insertion &&
2898 		    (flags & IBMVNIC_VLAN_STRIPPED))
2899 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2900 					       ntohs(next->rx_comp.vlan_tci));
2901 
2902 		/* free the entry */
2903 		next->rx_comp.first = 0;
2904 		remove_buff_from_pool(adapter, rx_buff);
2905 
2906 		skb_put(skb, length);
2907 		skb->protocol = eth_type_trans(skb, netdev);
2908 		skb_record_rx_queue(skb, scrq_num);
2909 
2910 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2911 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2912 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2913 		}
2914 
2915 		length = skb->len;
2916 		napi_gro_receive(napi, skb); /* send it up */
2917 		netdev->stats.rx_packets++;
2918 		netdev->stats.rx_bytes += length;
2919 		adapter->rx_stats_buffers[scrq_num].packets++;
2920 		adapter->rx_stats_buffers[scrq_num].bytes += length;
2921 		frames_processed++;
2922 	}
2923 
2924 	if (adapter->state != VNIC_CLOSING &&
2925 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2926 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
2927 	      frames_processed < budget))
2928 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2929 	if (frames_processed < budget) {
2930 		if (napi_complete_done(napi, frames_processed)) {
2931 			enable_scrq_irq(adapter, rx_scrq);
2932 			if (pending_scrq(adapter, rx_scrq)) {
2933 				if (napi_reschedule(napi)) {
2934 					disable_scrq_irq(adapter, rx_scrq);
2935 					goto restart_poll;
2936 				}
2937 			}
2938 		}
2939 	}
2940 	return frames_processed;
2941 }
2942 
2943 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2944 {
2945 	int rc, ret;
2946 
2947 	adapter->fallback.mtu = adapter->req_mtu;
2948 	adapter->fallback.rx_queues = adapter->req_rx_queues;
2949 	adapter->fallback.tx_queues = adapter->req_tx_queues;
2950 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2951 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2952 
2953 	reinit_completion(&adapter->reset_done);
2954 	adapter->wait_for_reset = true;
2955 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2956 
2957 	if (rc) {
2958 		ret = rc;
2959 		goto out;
2960 	}
2961 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2962 	if (rc) {
2963 		ret = -ENODEV;
2964 		goto out;
2965 	}
2966 
2967 	ret = 0;
2968 	if (adapter->reset_done_rc) {
2969 		ret = -EIO;
2970 		adapter->desired.mtu = adapter->fallback.mtu;
2971 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
2972 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
2973 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
2974 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
2975 
2976 		reinit_completion(&adapter->reset_done);
2977 		adapter->wait_for_reset = true;
2978 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2979 		if (rc) {
2980 			ret = rc;
2981 			goto out;
2982 		}
2983 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2984 						 60000);
2985 		if (rc) {
2986 			ret = -ENODEV;
2987 			goto out;
2988 		}
2989 	}
2990 out:
2991 	adapter->wait_for_reset = false;
2992 
2993 	return ret;
2994 }
2995 
2996 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2997 {
2998 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2999 
3000 	adapter->desired.mtu = new_mtu + ETH_HLEN;
3001 
3002 	return wait_for_reset(adapter);
3003 }
3004 
3005 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3006 						struct net_device *dev,
3007 						netdev_features_t features)
3008 {
3009 	/* Some backing hardware adapters can not
3010 	 * handle packets with a MSS less than 224
3011 	 * or with only one segment.
3012 	 */
3013 	if (skb_is_gso(skb)) {
3014 		if (skb_shinfo(skb)->gso_size < 224 ||
3015 		    skb_shinfo(skb)->gso_segs == 1)
3016 			features &= ~NETIF_F_GSO_MASK;
3017 	}
3018 
3019 	return features;
3020 }
3021 
3022 static const struct net_device_ops ibmvnic_netdev_ops = {
3023 	.ndo_open		= ibmvnic_open,
3024 	.ndo_stop		= ibmvnic_close,
3025 	.ndo_start_xmit		= ibmvnic_xmit,
3026 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3027 	.ndo_set_mac_address	= ibmvnic_set_mac,
3028 	.ndo_validate_addr	= eth_validate_addr,
3029 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3030 	.ndo_change_mtu		= ibmvnic_change_mtu,
3031 	.ndo_features_check     = ibmvnic_features_check,
3032 };
3033 
3034 /* ethtool functions */
3035 
3036 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3037 				      struct ethtool_link_ksettings *cmd)
3038 {
3039 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3040 	int rc;
3041 
3042 	rc = send_query_phys_parms(adapter);
3043 	if (rc) {
3044 		adapter->speed = SPEED_UNKNOWN;
3045 		adapter->duplex = DUPLEX_UNKNOWN;
3046 	}
3047 	cmd->base.speed = adapter->speed;
3048 	cmd->base.duplex = adapter->duplex;
3049 	cmd->base.port = PORT_FIBRE;
3050 	cmd->base.phy_address = 0;
3051 	cmd->base.autoneg = AUTONEG_ENABLE;
3052 
3053 	return 0;
3054 }
3055 
3056 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3057 				struct ethtool_drvinfo *info)
3058 {
3059 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3060 
3061 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3062 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3063 	strscpy(info->fw_version, adapter->fw_version,
3064 		sizeof(info->fw_version));
3065 }
3066 
3067 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3068 {
3069 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3070 
3071 	return adapter->msg_enable;
3072 }
3073 
3074 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3075 {
3076 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3077 
3078 	adapter->msg_enable = data;
3079 }
3080 
3081 static u32 ibmvnic_get_link(struct net_device *netdev)
3082 {
3083 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3084 
3085 	/* Don't need to send a query because we request a logical link up at
3086 	 * init and then we wait for link state indications
3087 	 */
3088 	return adapter->logical_link_state;
3089 }
3090 
3091 static void ibmvnic_get_ringparam(struct net_device *netdev,
3092 				  struct ethtool_ringparam *ring,
3093 				  struct kernel_ethtool_ringparam *kernel_ring,
3094 				  struct netlink_ext_ack *extack)
3095 {
3096 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3097 
3098 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3099 		ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3100 		ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3101 	} else {
3102 		ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3103 		ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3104 	}
3105 	ring->rx_mini_max_pending = 0;
3106 	ring->rx_jumbo_max_pending = 0;
3107 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3108 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3109 	ring->rx_mini_pending = 0;
3110 	ring->rx_jumbo_pending = 0;
3111 }
3112 
3113 static int ibmvnic_set_ringparam(struct net_device *netdev,
3114 				 struct ethtool_ringparam *ring,
3115 				 struct kernel_ethtool_ringparam *kernel_ring,
3116 				 struct netlink_ext_ack *extack)
3117 {
3118 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3119 	int ret;
3120 
3121 	ret = 0;
3122 	adapter->desired.rx_entries = ring->rx_pending;
3123 	adapter->desired.tx_entries = ring->tx_pending;
3124 
3125 	ret = wait_for_reset(adapter);
3126 
3127 	if (!ret &&
3128 	    (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
3129 	     adapter->req_tx_entries_per_subcrq != ring->tx_pending))
3130 		netdev_info(netdev,
3131 			    "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3132 			    ring->rx_pending, ring->tx_pending,
3133 			    adapter->req_rx_add_entries_per_subcrq,
3134 			    adapter->req_tx_entries_per_subcrq);
3135 	return ret;
3136 }
3137 
3138 static void ibmvnic_get_channels(struct net_device *netdev,
3139 				 struct ethtool_channels *channels)
3140 {
3141 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3142 
3143 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3144 		channels->max_rx = adapter->max_rx_queues;
3145 		channels->max_tx = adapter->max_tx_queues;
3146 	} else {
3147 		channels->max_rx = IBMVNIC_MAX_QUEUES;
3148 		channels->max_tx = IBMVNIC_MAX_QUEUES;
3149 	}
3150 
3151 	channels->max_other = 0;
3152 	channels->max_combined = 0;
3153 	channels->rx_count = adapter->req_rx_queues;
3154 	channels->tx_count = adapter->req_tx_queues;
3155 	channels->other_count = 0;
3156 	channels->combined_count = 0;
3157 }
3158 
3159 static int ibmvnic_set_channels(struct net_device *netdev,
3160 				struct ethtool_channels *channels)
3161 {
3162 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3163 	int ret;
3164 
3165 	ret = 0;
3166 	adapter->desired.rx_queues = channels->rx_count;
3167 	adapter->desired.tx_queues = channels->tx_count;
3168 
3169 	ret = wait_for_reset(adapter);
3170 
3171 	if (!ret &&
3172 	    (adapter->req_rx_queues != channels->rx_count ||
3173 	     adapter->req_tx_queues != channels->tx_count))
3174 		netdev_info(netdev,
3175 			    "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3176 			    channels->rx_count, channels->tx_count,
3177 			    adapter->req_rx_queues, adapter->req_tx_queues);
3178 	return ret;
3179 }
3180 
3181 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3182 {
3183 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3184 	int i;
3185 
3186 	switch (stringset) {
3187 	case ETH_SS_STATS:
3188 		for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
3189 				i++, data += ETH_GSTRING_LEN)
3190 			memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3191 
3192 		for (i = 0; i < adapter->req_tx_queues; i++) {
3193 			snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3194 			data += ETH_GSTRING_LEN;
3195 
3196 			snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3197 			data += ETH_GSTRING_LEN;
3198 
3199 			snprintf(data, ETH_GSTRING_LEN,
3200 				 "tx%d_dropped_packets", i);
3201 			data += ETH_GSTRING_LEN;
3202 		}
3203 
3204 		for (i = 0; i < adapter->req_rx_queues; i++) {
3205 			snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3206 			data += ETH_GSTRING_LEN;
3207 
3208 			snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3209 			data += ETH_GSTRING_LEN;
3210 
3211 			snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3212 			data += ETH_GSTRING_LEN;
3213 		}
3214 		break;
3215 
3216 	case ETH_SS_PRIV_FLAGS:
3217 		for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3218 			strcpy(data + i * ETH_GSTRING_LEN,
3219 			       ibmvnic_priv_flags[i]);
3220 		break;
3221 	default:
3222 		return;
3223 	}
3224 }
3225 
3226 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3227 {
3228 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3229 
3230 	switch (sset) {
3231 	case ETH_SS_STATS:
3232 		return ARRAY_SIZE(ibmvnic_stats) +
3233 		       adapter->req_tx_queues * NUM_TX_STATS +
3234 		       adapter->req_rx_queues * NUM_RX_STATS;
3235 	case ETH_SS_PRIV_FLAGS:
3236 		return ARRAY_SIZE(ibmvnic_priv_flags);
3237 	default:
3238 		return -EOPNOTSUPP;
3239 	}
3240 }
3241 
3242 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3243 				      struct ethtool_stats *stats, u64 *data)
3244 {
3245 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3246 	union ibmvnic_crq crq;
3247 	int i, j;
3248 	int rc;
3249 
3250 	memset(&crq, 0, sizeof(crq));
3251 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3252 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3253 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3254 	crq.request_statistics.len =
3255 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3256 
3257 	/* Wait for data to be written */
3258 	reinit_completion(&adapter->stats_done);
3259 	rc = ibmvnic_send_crq(adapter, &crq);
3260 	if (rc)
3261 		return;
3262 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3263 	if (rc)
3264 		return;
3265 
3266 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3267 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3268 				      (adapter, ibmvnic_stats[i].offset));
3269 
3270 	for (j = 0; j < adapter->req_tx_queues; j++) {
3271 		data[i] = adapter->tx_stats_buffers[j].packets;
3272 		i++;
3273 		data[i] = adapter->tx_stats_buffers[j].bytes;
3274 		i++;
3275 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3276 		i++;
3277 	}
3278 
3279 	for (j = 0; j < adapter->req_rx_queues; j++) {
3280 		data[i] = adapter->rx_stats_buffers[j].packets;
3281 		i++;
3282 		data[i] = adapter->rx_stats_buffers[j].bytes;
3283 		i++;
3284 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3285 		i++;
3286 	}
3287 }
3288 
3289 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3290 {
3291 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3292 
3293 	return adapter->priv_flags;
3294 }
3295 
3296 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3297 {
3298 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3299 	bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3300 
3301 	if (which_maxes)
3302 		adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3303 	else
3304 		adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3305 
3306 	return 0;
3307 }
3308 
3309 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3310 	.get_drvinfo		= ibmvnic_get_drvinfo,
3311 	.get_msglevel		= ibmvnic_get_msglevel,
3312 	.set_msglevel		= ibmvnic_set_msglevel,
3313 	.get_link		= ibmvnic_get_link,
3314 	.get_ringparam		= ibmvnic_get_ringparam,
3315 	.set_ringparam		= ibmvnic_set_ringparam,
3316 	.get_channels		= ibmvnic_get_channels,
3317 	.set_channels		= ibmvnic_set_channels,
3318 	.get_strings            = ibmvnic_get_strings,
3319 	.get_sset_count         = ibmvnic_get_sset_count,
3320 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3321 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3322 	.get_priv_flags		= ibmvnic_get_priv_flags,
3323 	.set_priv_flags		= ibmvnic_set_priv_flags,
3324 };
3325 
3326 /* Routines for managing CRQs/sCRQs  */
3327 
3328 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3329 				   struct ibmvnic_sub_crq_queue *scrq)
3330 {
3331 	int rc;
3332 
3333 	if (!scrq) {
3334 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3335 		return -EINVAL;
3336 	}
3337 
3338 	if (scrq->irq) {
3339 		free_irq(scrq->irq, scrq);
3340 		irq_dispose_mapping(scrq->irq);
3341 		scrq->irq = 0;
3342 	}
3343 
3344 	if (scrq->msgs) {
3345 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3346 		atomic_set(&scrq->used, 0);
3347 		scrq->cur = 0;
3348 		scrq->ind_buf.index = 0;
3349 	} else {
3350 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3351 		return -EINVAL;
3352 	}
3353 
3354 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3355 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3356 	return rc;
3357 }
3358 
3359 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3360 {
3361 	int i, rc;
3362 
3363 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3364 		return -EINVAL;
3365 
3366 	for (i = 0; i < adapter->req_tx_queues; i++) {
3367 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3368 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3369 		if (rc)
3370 			return rc;
3371 	}
3372 
3373 	for (i = 0; i < adapter->req_rx_queues; i++) {
3374 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3375 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3376 		if (rc)
3377 			return rc;
3378 	}
3379 
3380 	return rc;
3381 }
3382 
3383 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3384 				  struct ibmvnic_sub_crq_queue *scrq,
3385 				  bool do_h_free)
3386 {
3387 	struct device *dev = &adapter->vdev->dev;
3388 	long rc;
3389 
3390 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3391 
3392 	if (do_h_free) {
3393 		/* Close the sub-crqs */
3394 		do {
3395 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3396 						adapter->vdev->unit_address,
3397 						scrq->crq_num);
3398 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3399 
3400 		if (rc) {
3401 			netdev_err(adapter->netdev,
3402 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3403 				   scrq->crq_num, rc);
3404 		}
3405 	}
3406 
3407 	dma_free_coherent(dev,
3408 			  IBMVNIC_IND_ARR_SZ,
3409 			  scrq->ind_buf.indir_arr,
3410 			  scrq->ind_buf.indir_dma);
3411 
3412 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3413 			 DMA_BIDIRECTIONAL);
3414 	free_pages((unsigned long)scrq->msgs, 2);
3415 	kfree(scrq);
3416 }
3417 
3418 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3419 							*adapter)
3420 {
3421 	struct device *dev = &adapter->vdev->dev;
3422 	struct ibmvnic_sub_crq_queue *scrq;
3423 	int rc;
3424 
3425 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3426 	if (!scrq)
3427 		return NULL;
3428 
3429 	scrq->msgs =
3430 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3431 	if (!scrq->msgs) {
3432 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3433 		goto zero_page_failed;
3434 	}
3435 
3436 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3437 					 DMA_BIDIRECTIONAL);
3438 	if (dma_mapping_error(dev, scrq->msg_token)) {
3439 		dev_warn(dev, "Couldn't map crq queue messages page\n");
3440 		goto map_failed;
3441 	}
3442 
3443 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3444 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3445 
3446 	if (rc == H_RESOURCE)
3447 		rc = ibmvnic_reset_crq(adapter);
3448 
3449 	if (rc == H_CLOSED) {
3450 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
3451 	} else if (rc) {
3452 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
3453 		goto reg_failed;
3454 	}
3455 
3456 	scrq->adapter = adapter;
3457 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3458 	scrq->ind_buf.index = 0;
3459 
3460 	scrq->ind_buf.indir_arr =
3461 		dma_alloc_coherent(dev,
3462 				   IBMVNIC_IND_ARR_SZ,
3463 				   &scrq->ind_buf.indir_dma,
3464 				   GFP_KERNEL);
3465 
3466 	if (!scrq->ind_buf.indir_arr)
3467 		goto indir_failed;
3468 
3469 	spin_lock_init(&scrq->lock);
3470 
3471 	netdev_dbg(adapter->netdev,
3472 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3473 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
3474 
3475 	return scrq;
3476 
3477 indir_failed:
3478 	do {
3479 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3480 					adapter->vdev->unit_address,
3481 					scrq->crq_num);
3482 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3483 reg_failed:
3484 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3485 			 DMA_BIDIRECTIONAL);
3486 map_failed:
3487 	free_pages((unsigned long)scrq->msgs, 2);
3488 zero_page_failed:
3489 	kfree(scrq);
3490 
3491 	return NULL;
3492 }
3493 
3494 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3495 {
3496 	int i;
3497 
3498 	if (adapter->tx_scrq) {
3499 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3500 			if (!adapter->tx_scrq[i])
3501 				continue;
3502 
3503 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3504 				   i);
3505 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3506 			if (adapter->tx_scrq[i]->irq) {
3507 				free_irq(adapter->tx_scrq[i]->irq,
3508 					 adapter->tx_scrq[i]);
3509 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3510 				adapter->tx_scrq[i]->irq = 0;
3511 			}
3512 
3513 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3514 					      do_h_free);
3515 		}
3516 
3517 		kfree(adapter->tx_scrq);
3518 		adapter->tx_scrq = NULL;
3519 		adapter->num_active_tx_scrqs = 0;
3520 	}
3521 
3522 	if (adapter->rx_scrq) {
3523 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3524 			if (!adapter->rx_scrq[i])
3525 				continue;
3526 
3527 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3528 				   i);
3529 			if (adapter->rx_scrq[i]->irq) {
3530 				free_irq(adapter->rx_scrq[i]->irq,
3531 					 adapter->rx_scrq[i]);
3532 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3533 				adapter->rx_scrq[i]->irq = 0;
3534 			}
3535 
3536 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3537 					      do_h_free);
3538 		}
3539 
3540 		kfree(adapter->rx_scrq);
3541 		adapter->rx_scrq = NULL;
3542 		adapter->num_active_rx_scrqs = 0;
3543 	}
3544 }
3545 
3546 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3547 			    struct ibmvnic_sub_crq_queue *scrq)
3548 {
3549 	struct device *dev = &adapter->vdev->dev;
3550 	unsigned long rc;
3551 
3552 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3553 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3554 	if (rc)
3555 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3556 			scrq->hw_irq, rc);
3557 	return rc;
3558 }
3559 
3560 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3561 			   struct ibmvnic_sub_crq_queue *scrq)
3562 {
3563 	struct device *dev = &adapter->vdev->dev;
3564 	unsigned long rc;
3565 
3566 	if (scrq->hw_irq > 0x100000000ULL) {
3567 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3568 		return 1;
3569 	}
3570 
3571 	if (test_bit(0, &adapter->resetting) &&
3572 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
3573 		u64 val = (0xff000000) | scrq->hw_irq;
3574 
3575 		rc = plpar_hcall_norets(H_EOI, val);
3576 		/* H_EOI would fail with rc = H_FUNCTION when running
3577 		 * in XIVE mode which is expected, but not an error.
3578 		 */
3579 		if (rc && (rc != H_FUNCTION))
3580 			dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3581 				val, rc);
3582 	}
3583 
3584 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3585 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3586 	if (rc)
3587 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3588 			scrq->hw_irq, rc);
3589 	return rc;
3590 }
3591 
3592 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3593 			       struct ibmvnic_sub_crq_queue *scrq)
3594 {
3595 	struct device *dev = &adapter->vdev->dev;
3596 	struct ibmvnic_tx_pool *tx_pool;
3597 	struct ibmvnic_tx_buff *txbuff;
3598 	struct netdev_queue *txq;
3599 	union sub_crq *next;
3600 	int index;
3601 	int i;
3602 
3603 restart_loop:
3604 	while (pending_scrq(adapter, scrq)) {
3605 		unsigned int pool = scrq->pool_index;
3606 		int num_entries = 0;
3607 		int total_bytes = 0;
3608 		int num_packets = 0;
3609 
3610 		next = ibmvnic_next_scrq(adapter, scrq);
3611 		for (i = 0; i < next->tx_comp.num_comps; i++) {
3612 			index = be32_to_cpu(next->tx_comp.correlators[i]);
3613 			if (index & IBMVNIC_TSO_POOL_MASK) {
3614 				tx_pool = &adapter->tso_pool[pool];
3615 				index &= ~IBMVNIC_TSO_POOL_MASK;
3616 			} else {
3617 				tx_pool = &adapter->tx_pool[pool];
3618 			}
3619 
3620 			txbuff = &tx_pool->tx_buff[index];
3621 			num_packets++;
3622 			num_entries += txbuff->num_entries;
3623 			if (txbuff->skb) {
3624 				total_bytes += txbuff->skb->len;
3625 				if (next->tx_comp.rcs[i]) {
3626 					dev_err(dev, "tx error %x\n",
3627 						next->tx_comp.rcs[i]);
3628 					dev_kfree_skb_irq(txbuff->skb);
3629 				} else {
3630 					dev_consume_skb_irq(txbuff->skb);
3631 				}
3632 				txbuff->skb = NULL;
3633 			} else {
3634 				netdev_warn(adapter->netdev,
3635 					    "TX completion received with NULL socket buffer\n");
3636 			}
3637 			tx_pool->free_map[tx_pool->producer_index] = index;
3638 			tx_pool->producer_index =
3639 				(tx_pool->producer_index + 1) %
3640 					tx_pool->num_buffers;
3641 		}
3642 		/* remove tx_comp scrq*/
3643 		next->tx_comp.first = 0;
3644 
3645 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3646 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
3647 
3648 		if (atomic_sub_return(num_entries, &scrq->used) <=
3649 		    (adapter->req_tx_entries_per_subcrq / 2) &&
3650 		    __netif_subqueue_stopped(adapter->netdev,
3651 					     scrq->pool_index)) {
3652 			netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3653 			netdev_dbg(adapter->netdev, "Started queue %d\n",
3654 				   scrq->pool_index);
3655 		}
3656 	}
3657 
3658 	enable_scrq_irq(adapter, scrq);
3659 
3660 	if (pending_scrq(adapter, scrq)) {
3661 		disable_scrq_irq(adapter, scrq);
3662 		goto restart_loop;
3663 	}
3664 
3665 	return 0;
3666 }
3667 
3668 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3669 {
3670 	struct ibmvnic_sub_crq_queue *scrq = instance;
3671 	struct ibmvnic_adapter *adapter = scrq->adapter;
3672 
3673 	disable_scrq_irq(adapter, scrq);
3674 	ibmvnic_complete_tx(adapter, scrq);
3675 
3676 	return IRQ_HANDLED;
3677 }
3678 
3679 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3680 {
3681 	struct ibmvnic_sub_crq_queue *scrq = instance;
3682 	struct ibmvnic_adapter *adapter = scrq->adapter;
3683 
3684 	/* When booting a kdump kernel we can hit pending interrupts
3685 	 * prior to completing driver initialization.
3686 	 */
3687 	if (unlikely(adapter->state != VNIC_OPEN))
3688 		return IRQ_NONE;
3689 
3690 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3691 
3692 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3693 		disable_scrq_irq(adapter, scrq);
3694 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
3695 	}
3696 
3697 	return IRQ_HANDLED;
3698 }
3699 
3700 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3701 {
3702 	struct device *dev = &adapter->vdev->dev;
3703 	struct ibmvnic_sub_crq_queue *scrq;
3704 	int i = 0, j = 0;
3705 	int rc = 0;
3706 
3707 	for (i = 0; i < adapter->req_tx_queues; i++) {
3708 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3709 			   i);
3710 		scrq = adapter->tx_scrq[i];
3711 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3712 
3713 		if (!scrq->irq) {
3714 			rc = -EINVAL;
3715 			dev_err(dev, "Error mapping irq\n");
3716 			goto req_tx_irq_failed;
3717 		}
3718 
3719 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3720 			 adapter->vdev->unit_address, i);
3721 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3722 				 0, scrq->name, scrq);
3723 
3724 		if (rc) {
3725 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3726 				scrq->irq, rc);
3727 			irq_dispose_mapping(scrq->irq);
3728 			goto req_tx_irq_failed;
3729 		}
3730 	}
3731 
3732 	for (i = 0; i < adapter->req_rx_queues; i++) {
3733 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3734 			   i);
3735 		scrq = adapter->rx_scrq[i];
3736 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3737 		if (!scrq->irq) {
3738 			rc = -EINVAL;
3739 			dev_err(dev, "Error mapping irq\n");
3740 			goto req_rx_irq_failed;
3741 		}
3742 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3743 			 adapter->vdev->unit_address, i);
3744 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3745 				 0, scrq->name, scrq);
3746 		if (rc) {
3747 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3748 				scrq->irq, rc);
3749 			irq_dispose_mapping(scrq->irq);
3750 			goto req_rx_irq_failed;
3751 		}
3752 	}
3753 	return rc;
3754 
3755 req_rx_irq_failed:
3756 	for (j = 0; j < i; j++) {
3757 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3758 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3759 	}
3760 	i = adapter->req_tx_queues;
3761 req_tx_irq_failed:
3762 	for (j = 0; j < i; j++) {
3763 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3764 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3765 	}
3766 	release_sub_crqs(adapter, 1);
3767 	return rc;
3768 }
3769 
3770 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3771 {
3772 	struct device *dev = &adapter->vdev->dev;
3773 	struct ibmvnic_sub_crq_queue **allqueues;
3774 	int registered_queues = 0;
3775 	int total_queues;
3776 	int more = 0;
3777 	int i;
3778 
3779 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3780 
3781 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3782 	if (!allqueues)
3783 		return -ENOMEM;
3784 
3785 	for (i = 0; i < total_queues; i++) {
3786 		allqueues[i] = init_sub_crq_queue(adapter);
3787 		if (!allqueues[i]) {
3788 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3789 			break;
3790 		}
3791 		registered_queues++;
3792 	}
3793 
3794 	/* Make sure we were able to register the minimum number of queues */
3795 	if (registered_queues <
3796 	    adapter->min_tx_queues + adapter->min_rx_queues) {
3797 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3798 		goto tx_failed;
3799 	}
3800 
3801 	/* Distribute the failed allocated queues*/
3802 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
3803 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3804 		switch (i % 3) {
3805 		case 0:
3806 			if (adapter->req_rx_queues > adapter->min_rx_queues)
3807 				adapter->req_rx_queues--;
3808 			else
3809 				more++;
3810 			break;
3811 		case 1:
3812 			if (adapter->req_tx_queues > adapter->min_tx_queues)
3813 				adapter->req_tx_queues--;
3814 			else
3815 				more++;
3816 			break;
3817 		}
3818 	}
3819 
3820 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3821 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
3822 	if (!adapter->tx_scrq)
3823 		goto tx_failed;
3824 
3825 	for (i = 0; i < adapter->req_tx_queues; i++) {
3826 		adapter->tx_scrq[i] = allqueues[i];
3827 		adapter->tx_scrq[i]->pool_index = i;
3828 		adapter->num_active_tx_scrqs++;
3829 	}
3830 
3831 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3832 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
3833 	if (!adapter->rx_scrq)
3834 		goto rx_failed;
3835 
3836 	for (i = 0; i < adapter->req_rx_queues; i++) {
3837 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3838 		adapter->rx_scrq[i]->scrq_num = i;
3839 		adapter->num_active_rx_scrqs++;
3840 	}
3841 
3842 	kfree(allqueues);
3843 	return 0;
3844 
3845 rx_failed:
3846 	kfree(adapter->tx_scrq);
3847 	adapter->tx_scrq = NULL;
3848 tx_failed:
3849 	for (i = 0; i < registered_queues; i++)
3850 		release_sub_crq_queue(adapter, allqueues[i], 1);
3851 	kfree(allqueues);
3852 	return -ENOMEM;
3853 }
3854 
3855 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3856 {
3857 	struct device *dev = &adapter->vdev->dev;
3858 	union ibmvnic_crq crq;
3859 	int max_entries;
3860 	int cap_reqs;
3861 
3862 	/* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
3863 	 * the PROMISC flag). Initialize this count upfront. When the tasklet
3864 	 * receives a response to all of these, it will send the next protocol
3865 	 * message (QUERY_IP_OFFLOAD).
3866 	 */
3867 	if (!(adapter->netdev->flags & IFF_PROMISC) ||
3868 	    adapter->promisc_supported)
3869 		cap_reqs = 7;
3870 	else
3871 		cap_reqs = 6;
3872 
3873 	if (!retry) {
3874 		/* Sub-CRQ entries are 32 byte long */
3875 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3876 
3877 		atomic_set(&adapter->running_cap_crqs, cap_reqs);
3878 
3879 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
3880 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
3881 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3882 			return;
3883 		}
3884 
3885 		if (adapter->desired.mtu)
3886 			adapter->req_mtu = adapter->desired.mtu;
3887 		else
3888 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3889 
3890 		if (!adapter->desired.tx_entries)
3891 			adapter->desired.tx_entries =
3892 					adapter->max_tx_entries_per_subcrq;
3893 		if (!adapter->desired.rx_entries)
3894 			adapter->desired.rx_entries =
3895 					adapter->max_rx_add_entries_per_subcrq;
3896 
3897 		max_entries = IBMVNIC_MAX_LTB_SIZE /
3898 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3899 
3900 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3901 			adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3902 			adapter->desired.tx_entries = max_entries;
3903 		}
3904 
3905 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3906 			adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3907 			adapter->desired.rx_entries = max_entries;
3908 		}
3909 
3910 		if (adapter->desired.tx_entries)
3911 			adapter->req_tx_entries_per_subcrq =
3912 					adapter->desired.tx_entries;
3913 		else
3914 			adapter->req_tx_entries_per_subcrq =
3915 					adapter->max_tx_entries_per_subcrq;
3916 
3917 		if (adapter->desired.rx_entries)
3918 			adapter->req_rx_add_entries_per_subcrq =
3919 					adapter->desired.rx_entries;
3920 		else
3921 			adapter->req_rx_add_entries_per_subcrq =
3922 					adapter->max_rx_add_entries_per_subcrq;
3923 
3924 		if (adapter->desired.tx_queues)
3925 			adapter->req_tx_queues =
3926 					adapter->desired.tx_queues;
3927 		else
3928 			adapter->req_tx_queues =
3929 					adapter->opt_tx_comp_sub_queues;
3930 
3931 		if (adapter->desired.rx_queues)
3932 			adapter->req_rx_queues =
3933 					adapter->desired.rx_queues;
3934 		else
3935 			adapter->req_rx_queues =
3936 					adapter->opt_rx_comp_queues;
3937 
3938 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3939 	} else {
3940 		atomic_add(cap_reqs, &adapter->running_cap_crqs);
3941 	}
3942 	memset(&crq, 0, sizeof(crq));
3943 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
3944 	crq.request_capability.cmd = REQUEST_CAPABILITY;
3945 
3946 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3947 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3948 	cap_reqs--;
3949 	ibmvnic_send_crq(adapter, &crq);
3950 
3951 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3952 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3953 	cap_reqs--;
3954 	ibmvnic_send_crq(adapter, &crq);
3955 
3956 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3957 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3958 	cap_reqs--;
3959 	ibmvnic_send_crq(adapter, &crq);
3960 
3961 	crq.request_capability.capability =
3962 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3963 	crq.request_capability.number =
3964 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3965 	cap_reqs--;
3966 	ibmvnic_send_crq(adapter, &crq);
3967 
3968 	crq.request_capability.capability =
3969 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3970 	crq.request_capability.number =
3971 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3972 	cap_reqs--;
3973 	ibmvnic_send_crq(adapter, &crq);
3974 
3975 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3976 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3977 	cap_reqs--;
3978 	ibmvnic_send_crq(adapter, &crq);
3979 
3980 	if (adapter->netdev->flags & IFF_PROMISC) {
3981 		if (adapter->promisc_supported) {
3982 			crq.request_capability.capability =
3983 			    cpu_to_be16(PROMISC_REQUESTED);
3984 			crq.request_capability.number = cpu_to_be64(1);
3985 			cap_reqs--;
3986 			ibmvnic_send_crq(adapter, &crq);
3987 		}
3988 	} else {
3989 		crq.request_capability.capability =
3990 		    cpu_to_be16(PROMISC_REQUESTED);
3991 		crq.request_capability.number = cpu_to_be64(0);
3992 		cap_reqs--;
3993 		ibmvnic_send_crq(adapter, &crq);
3994 	}
3995 
3996 	/* Keep at end to catch any discrepancy between expected and actual
3997 	 * CRQs sent.
3998 	 */
3999 	WARN_ON(cap_reqs != 0);
4000 }
4001 
4002 static int pending_scrq(struct ibmvnic_adapter *adapter,
4003 			struct ibmvnic_sub_crq_queue *scrq)
4004 {
4005 	union sub_crq *entry = &scrq->msgs[scrq->cur];
4006 	int rc;
4007 
4008 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4009 
4010 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4011 	 * contents of the SCRQ descriptor
4012 	 */
4013 	dma_rmb();
4014 
4015 	return rc;
4016 }
4017 
4018 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4019 					struct ibmvnic_sub_crq_queue *scrq)
4020 {
4021 	union sub_crq *entry;
4022 	unsigned long flags;
4023 
4024 	spin_lock_irqsave(&scrq->lock, flags);
4025 	entry = &scrq->msgs[scrq->cur];
4026 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4027 		if (++scrq->cur == scrq->size)
4028 			scrq->cur = 0;
4029 	} else {
4030 		entry = NULL;
4031 	}
4032 	spin_unlock_irqrestore(&scrq->lock, flags);
4033 
4034 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4035 	 * contents of the SCRQ descriptor
4036 	 */
4037 	dma_rmb();
4038 
4039 	return entry;
4040 }
4041 
4042 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4043 {
4044 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4045 	union ibmvnic_crq *crq;
4046 
4047 	crq = &queue->msgs[queue->cur];
4048 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4049 		if (++queue->cur == queue->size)
4050 			queue->cur = 0;
4051 	} else {
4052 		crq = NULL;
4053 	}
4054 
4055 	return crq;
4056 }
4057 
4058 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4059 {
4060 	switch (rc) {
4061 	case H_PARAMETER:
4062 		dev_warn_ratelimited(dev,
4063 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4064 				     func, rc);
4065 		break;
4066 	case H_CLOSED:
4067 		dev_warn_ratelimited(dev,
4068 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4069 				     func, rc);
4070 		break;
4071 	default:
4072 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4073 		break;
4074 	}
4075 }
4076 
4077 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4078 				u64 remote_handle, u64 ioba, u64 num_entries)
4079 {
4080 	unsigned int ua = adapter->vdev->unit_address;
4081 	struct device *dev = &adapter->vdev->dev;
4082 	int rc;
4083 
4084 	/* Make sure the hypervisor sees the complete request */
4085 	dma_wmb();
4086 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4087 				cpu_to_be64(remote_handle),
4088 				ioba, num_entries);
4089 
4090 	if (rc)
4091 		print_subcrq_error(dev, rc, __func__);
4092 
4093 	return rc;
4094 }
4095 
4096 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4097 			    union ibmvnic_crq *crq)
4098 {
4099 	unsigned int ua = adapter->vdev->unit_address;
4100 	struct device *dev = &adapter->vdev->dev;
4101 	u64 *u64_crq = (u64 *)crq;
4102 	int rc;
4103 
4104 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4105 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4106 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4107 
4108 	if (!adapter->crq.active &&
4109 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4110 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4111 		return -EINVAL;
4112 	}
4113 
4114 	/* Make sure the hypervisor sees the complete request */
4115 	dma_wmb();
4116 
4117 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4118 				cpu_to_be64(u64_crq[0]),
4119 				cpu_to_be64(u64_crq[1]));
4120 
4121 	if (rc) {
4122 		if (rc == H_CLOSED) {
4123 			dev_warn(dev, "CRQ Queue closed\n");
4124 			/* do not reset, report the fail, wait for passive init from server */
4125 		}
4126 
4127 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4128 	}
4129 
4130 	return rc;
4131 }
4132 
4133 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4134 {
4135 	struct device *dev = &adapter->vdev->dev;
4136 	union ibmvnic_crq crq;
4137 	int retries = 100;
4138 	int rc;
4139 
4140 	memset(&crq, 0, sizeof(crq));
4141 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4142 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4143 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4144 
4145 	do {
4146 		rc = ibmvnic_send_crq(adapter, &crq);
4147 		if (rc != H_CLOSED)
4148 			break;
4149 		retries--;
4150 		msleep(50);
4151 
4152 	} while (retries > 0);
4153 
4154 	if (rc) {
4155 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4156 		return rc;
4157 	}
4158 
4159 	return 0;
4160 }
4161 
4162 struct vnic_login_client_data {
4163 	u8	type;
4164 	__be16	len;
4165 	char	name[];
4166 } __packed;
4167 
4168 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4169 {
4170 	int len;
4171 
4172 	/* Calculate the amount of buffer space needed for the
4173 	 * vnic client data in the login buffer. There are four entries,
4174 	 * OS name, LPAR name, device name, and a null last entry.
4175 	 */
4176 	len = 4 * sizeof(struct vnic_login_client_data);
4177 	len += 6; /* "Linux" plus NULL */
4178 	len += strlen(utsname()->nodename) + 1;
4179 	len += strlen(adapter->netdev->name) + 1;
4180 
4181 	return len;
4182 }
4183 
4184 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4185 				 struct vnic_login_client_data *vlcd)
4186 {
4187 	const char *os_name = "Linux";
4188 	int len;
4189 
4190 	/* Type 1 - LPAR OS */
4191 	vlcd->type = 1;
4192 	len = strlen(os_name) + 1;
4193 	vlcd->len = cpu_to_be16(len);
4194 	strscpy(vlcd->name, os_name, len);
4195 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4196 
4197 	/* Type 2 - LPAR name */
4198 	vlcd->type = 2;
4199 	len = strlen(utsname()->nodename) + 1;
4200 	vlcd->len = cpu_to_be16(len);
4201 	strscpy(vlcd->name, utsname()->nodename, len);
4202 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4203 
4204 	/* Type 3 - device name */
4205 	vlcd->type = 3;
4206 	len = strlen(adapter->netdev->name) + 1;
4207 	vlcd->len = cpu_to_be16(len);
4208 	strscpy(vlcd->name, adapter->netdev->name, len);
4209 }
4210 
4211 static int send_login(struct ibmvnic_adapter *adapter)
4212 {
4213 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4214 	struct ibmvnic_login_buffer *login_buffer;
4215 	struct device *dev = &adapter->vdev->dev;
4216 	struct vnic_login_client_data *vlcd;
4217 	dma_addr_t rsp_buffer_token;
4218 	dma_addr_t buffer_token;
4219 	size_t rsp_buffer_size;
4220 	union ibmvnic_crq crq;
4221 	int client_data_len;
4222 	size_t buffer_size;
4223 	__be64 *tx_list_p;
4224 	__be64 *rx_list_p;
4225 	int rc;
4226 	int i;
4227 
4228 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4229 		netdev_err(adapter->netdev,
4230 			   "RX or TX queues are not allocated, device login failed\n");
4231 		return -ENOMEM;
4232 	}
4233 
4234 	release_login_buffer(adapter);
4235 	release_login_rsp_buffer(adapter);
4236 
4237 	client_data_len = vnic_client_data_len(adapter);
4238 
4239 	buffer_size =
4240 	    sizeof(struct ibmvnic_login_buffer) +
4241 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4242 	    client_data_len;
4243 
4244 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4245 	if (!login_buffer)
4246 		goto buf_alloc_failed;
4247 
4248 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4249 				      DMA_TO_DEVICE);
4250 	if (dma_mapping_error(dev, buffer_token)) {
4251 		dev_err(dev, "Couldn't map login buffer\n");
4252 		goto buf_map_failed;
4253 	}
4254 
4255 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4256 			  sizeof(u64) * adapter->req_tx_queues +
4257 			  sizeof(u64) * adapter->req_rx_queues +
4258 			  sizeof(u64) * adapter->req_rx_queues +
4259 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4260 
4261 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4262 	if (!login_rsp_buffer)
4263 		goto buf_rsp_alloc_failed;
4264 
4265 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4266 					  rsp_buffer_size, DMA_FROM_DEVICE);
4267 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4268 		dev_err(dev, "Couldn't map login rsp buffer\n");
4269 		goto buf_rsp_map_failed;
4270 	}
4271 
4272 	adapter->login_buf = login_buffer;
4273 	adapter->login_buf_token = buffer_token;
4274 	adapter->login_buf_sz = buffer_size;
4275 	adapter->login_rsp_buf = login_rsp_buffer;
4276 	adapter->login_rsp_buf_token = rsp_buffer_token;
4277 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4278 
4279 	login_buffer->len = cpu_to_be32(buffer_size);
4280 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4281 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4282 	login_buffer->off_txcomp_subcrqs =
4283 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4284 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4285 	login_buffer->off_rxcomp_subcrqs =
4286 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4287 			sizeof(u64) * adapter->req_tx_queues);
4288 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4289 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4290 
4291 	tx_list_p = (__be64 *)((char *)login_buffer +
4292 				      sizeof(struct ibmvnic_login_buffer));
4293 	rx_list_p = (__be64 *)((char *)login_buffer +
4294 				      sizeof(struct ibmvnic_login_buffer) +
4295 				      sizeof(u64) * adapter->req_tx_queues);
4296 
4297 	for (i = 0; i < adapter->req_tx_queues; i++) {
4298 		if (adapter->tx_scrq[i]) {
4299 			tx_list_p[i] =
4300 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4301 		}
4302 	}
4303 
4304 	for (i = 0; i < adapter->req_rx_queues; i++) {
4305 		if (adapter->rx_scrq[i]) {
4306 			rx_list_p[i] =
4307 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4308 		}
4309 	}
4310 
4311 	/* Insert vNIC login client data */
4312 	vlcd = (struct vnic_login_client_data *)
4313 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4314 	login_buffer->client_data_offset =
4315 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4316 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4317 
4318 	vnic_add_client_data(adapter, vlcd);
4319 
4320 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4321 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4322 		netdev_dbg(adapter->netdev, "%016lx\n",
4323 			   ((unsigned long *)(adapter->login_buf))[i]);
4324 	}
4325 
4326 	memset(&crq, 0, sizeof(crq));
4327 	crq.login.first = IBMVNIC_CRQ_CMD;
4328 	crq.login.cmd = LOGIN;
4329 	crq.login.ioba = cpu_to_be32(buffer_token);
4330 	crq.login.len = cpu_to_be32(buffer_size);
4331 
4332 	adapter->login_pending = true;
4333 	rc = ibmvnic_send_crq(adapter, &crq);
4334 	if (rc) {
4335 		adapter->login_pending = false;
4336 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4337 		goto buf_rsp_map_failed;
4338 	}
4339 
4340 	return 0;
4341 
4342 buf_rsp_map_failed:
4343 	kfree(login_rsp_buffer);
4344 	adapter->login_rsp_buf = NULL;
4345 buf_rsp_alloc_failed:
4346 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4347 buf_map_failed:
4348 	kfree(login_buffer);
4349 	adapter->login_buf = NULL;
4350 buf_alloc_failed:
4351 	return -ENOMEM;
4352 }
4353 
4354 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4355 			    u32 len, u8 map_id)
4356 {
4357 	union ibmvnic_crq crq;
4358 
4359 	memset(&crq, 0, sizeof(crq));
4360 	crq.request_map.first = IBMVNIC_CRQ_CMD;
4361 	crq.request_map.cmd = REQUEST_MAP;
4362 	crq.request_map.map_id = map_id;
4363 	crq.request_map.ioba = cpu_to_be32(addr);
4364 	crq.request_map.len = cpu_to_be32(len);
4365 	return ibmvnic_send_crq(adapter, &crq);
4366 }
4367 
4368 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4369 {
4370 	union ibmvnic_crq crq;
4371 
4372 	memset(&crq, 0, sizeof(crq));
4373 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4374 	crq.request_unmap.cmd = REQUEST_UNMAP;
4375 	crq.request_unmap.map_id = map_id;
4376 	return ibmvnic_send_crq(adapter, &crq);
4377 }
4378 
4379 static void send_query_map(struct ibmvnic_adapter *adapter)
4380 {
4381 	union ibmvnic_crq crq;
4382 
4383 	memset(&crq, 0, sizeof(crq));
4384 	crq.query_map.first = IBMVNIC_CRQ_CMD;
4385 	crq.query_map.cmd = QUERY_MAP;
4386 	ibmvnic_send_crq(adapter, &crq);
4387 }
4388 
4389 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4390 static void send_query_cap(struct ibmvnic_adapter *adapter)
4391 {
4392 	union ibmvnic_crq crq;
4393 	int cap_reqs;
4394 
4395 	/* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
4396 	 * upfront. When the tasklet receives a response to all of these, it
4397 	 * can send out the next protocol messaage (REQUEST_CAPABILITY).
4398 	 */
4399 	cap_reqs = 25;
4400 
4401 	atomic_set(&adapter->running_cap_crqs, cap_reqs);
4402 
4403 	memset(&crq, 0, sizeof(crq));
4404 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
4405 	crq.query_capability.cmd = QUERY_CAPABILITY;
4406 
4407 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4408 	ibmvnic_send_crq(adapter, &crq);
4409 	cap_reqs--;
4410 
4411 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4412 	ibmvnic_send_crq(adapter, &crq);
4413 	cap_reqs--;
4414 
4415 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4416 	ibmvnic_send_crq(adapter, &crq);
4417 	cap_reqs--;
4418 
4419 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4420 	ibmvnic_send_crq(adapter, &crq);
4421 	cap_reqs--;
4422 
4423 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4424 	ibmvnic_send_crq(adapter, &crq);
4425 	cap_reqs--;
4426 
4427 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4428 	ibmvnic_send_crq(adapter, &crq);
4429 	cap_reqs--;
4430 
4431 	crq.query_capability.capability =
4432 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4433 	ibmvnic_send_crq(adapter, &crq);
4434 	cap_reqs--;
4435 
4436 	crq.query_capability.capability =
4437 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4438 	ibmvnic_send_crq(adapter, &crq);
4439 	cap_reqs--;
4440 
4441 	crq.query_capability.capability =
4442 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4443 	ibmvnic_send_crq(adapter, &crq);
4444 	cap_reqs--;
4445 
4446 	crq.query_capability.capability =
4447 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4448 	ibmvnic_send_crq(adapter, &crq);
4449 	cap_reqs--;
4450 
4451 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4452 	ibmvnic_send_crq(adapter, &crq);
4453 	cap_reqs--;
4454 
4455 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4456 	ibmvnic_send_crq(adapter, &crq);
4457 	cap_reqs--;
4458 
4459 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4460 	ibmvnic_send_crq(adapter, &crq);
4461 	cap_reqs--;
4462 
4463 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4464 	ibmvnic_send_crq(adapter, &crq);
4465 	cap_reqs--;
4466 
4467 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4468 	ibmvnic_send_crq(adapter, &crq);
4469 	cap_reqs--;
4470 
4471 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4472 	ibmvnic_send_crq(adapter, &crq);
4473 	cap_reqs--;
4474 
4475 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4476 	ibmvnic_send_crq(adapter, &crq);
4477 	cap_reqs--;
4478 
4479 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4480 	ibmvnic_send_crq(adapter, &crq);
4481 	cap_reqs--;
4482 
4483 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4484 	ibmvnic_send_crq(adapter, &crq);
4485 	cap_reqs--;
4486 
4487 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4488 	ibmvnic_send_crq(adapter, &crq);
4489 	cap_reqs--;
4490 
4491 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4492 	ibmvnic_send_crq(adapter, &crq);
4493 	cap_reqs--;
4494 
4495 	crq.query_capability.capability =
4496 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4497 	ibmvnic_send_crq(adapter, &crq);
4498 	cap_reqs--;
4499 
4500 	crq.query_capability.capability =
4501 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4502 	ibmvnic_send_crq(adapter, &crq);
4503 	cap_reqs--;
4504 
4505 	crq.query_capability.capability =
4506 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4507 	ibmvnic_send_crq(adapter, &crq);
4508 	cap_reqs--;
4509 
4510 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4511 
4512 	ibmvnic_send_crq(adapter, &crq);
4513 	cap_reqs--;
4514 
4515 	/* Keep at end to catch any discrepancy between expected and actual
4516 	 * CRQs sent.
4517 	 */
4518 	WARN_ON(cap_reqs != 0);
4519 }
4520 
4521 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4522 {
4523 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4524 	struct device *dev = &adapter->vdev->dev;
4525 	union ibmvnic_crq crq;
4526 
4527 	adapter->ip_offload_tok =
4528 		dma_map_single(dev,
4529 			       &adapter->ip_offload_buf,
4530 			       buf_sz,
4531 			       DMA_FROM_DEVICE);
4532 
4533 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4534 		if (!firmware_has_feature(FW_FEATURE_CMO))
4535 			dev_err(dev, "Couldn't map offload buffer\n");
4536 		return;
4537 	}
4538 
4539 	memset(&crq, 0, sizeof(crq));
4540 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4541 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4542 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4543 	crq.query_ip_offload.ioba =
4544 	    cpu_to_be32(adapter->ip_offload_tok);
4545 
4546 	ibmvnic_send_crq(adapter, &crq);
4547 }
4548 
4549 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4550 {
4551 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4552 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4553 	struct device *dev = &adapter->vdev->dev;
4554 	netdev_features_t old_hw_features = 0;
4555 	union ibmvnic_crq crq;
4556 
4557 	adapter->ip_offload_ctrl_tok =
4558 		dma_map_single(dev,
4559 			       ctrl_buf,
4560 			       sizeof(adapter->ip_offload_ctrl),
4561 			       DMA_TO_DEVICE);
4562 
4563 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4564 		dev_err(dev, "Couldn't map ip offload control buffer\n");
4565 		return;
4566 	}
4567 
4568 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4569 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4570 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4571 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4572 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4573 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4574 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4575 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4576 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4577 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4578 
4579 	/* large_rx disabled for now, additional features needed */
4580 	ctrl_buf->large_rx_ipv4 = 0;
4581 	ctrl_buf->large_rx_ipv6 = 0;
4582 
4583 	if (adapter->state != VNIC_PROBING) {
4584 		old_hw_features = adapter->netdev->hw_features;
4585 		adapter->netdev->hw_features = 0;
4586 	}
4587 
4588 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4589 
4590 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4591 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4592 
4593 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4594 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4595 
4596 	if ((adapter->netdev->features &
4597 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4598 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4599 
4600 	if (buf->large_tx_ipv4)
4601 		adapter->netdev->hw_features |= NETIF_F_TSO;
4602 	if (buf->large_tx_ipv6)
4603 		adapter->netdev->hw_features |= NETIF_F_TSO6;
4604 
4605 	if (adapter->state == VNIC_PROBING) {
4606 		adapter->netdev->features |= adapter->netdev->hw_features;
4607 	} else if (old_hw_features != adapter->netdev->hw_features) {
4608 		netdev_features_t tmp = 0;
4609 
4610 		/* disable features no longer supported */
4611 		adapter->netdev->features &= adapter->netdev->hw_features;
4612 		/* turn on features now supported if previously enabled */
4613 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4614 			adapter->netdev->hw_features;
4615 		adapter->netdev->features |=
4616 				tmp & adapter->netdev->wanted_features;
4617 	}
4618 
4619 	memset(&crq, 0, sizeof(crq));
4620 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4621 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4622 	crq.control_ip_offload.len =
4623 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4624 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4625 	ibmvnic_send_crq(adapter, &crq);
4626 }
4627 
4628 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4629 				struct ibmvnic_adapter *adapter)
4630 {
4631 	struct device *dev = &adapter->vdev->dev;
4632 
4633 	if (crq->get_vpd_size_rsp.rc.code) {
4634 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4635 			crq->get_vpd_size_rsp.rc.code);
4636 		complete(&adapter->fw_done);
4637 		return;
4638 	}
4639 
4640 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4641 	complete(&adapter->fw_done);
4642 }
4643 
4644 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4645 			   struct ibmvnic_adapter *adapter)
4646 {
4647 	struct device *dev = &adapter->vdev->dev;
4648 	unsigned char *substr = NULL;
4649 	u8 fw_level_len = 0;
4650 
4651 	memset(adapter->fw_version, 0, 32);
4652 
4653 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4654 			 DMA_FROM_DEVICE);
4655 
4656 	if (crq->get_vpd_rsp.rc.code) {
4657 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4658 			crq->get_vpd_rsp.rc.code);
4659 		goto complete;
4660 	}
4661 
4662 	/* get the position of the firmware version info
4663 	 * located after the ASCII 'RM' substring in the buffer
4664 	 */
4665 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4666 	if (!substr) {
4667 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4668 		goto complete;
4669 	}
4670 
4671 	/* get length of firmware level ASCII substring */
4672 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4673 		fw_level_len = *(substr + 2);
4674 	} else {
4675 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4676 		goto complete;
4677 	}
4678 
4679 	/* copy firmware version string from vpd into adapter */
4680 	if ((substr + 3 + fw_level_len) <
4681 	    (adapter->vpd->buff + adapter->vpd->len)) {
4682 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4683 	} else {
4684 		dev_info(dev, "FW substr extrapolated VPD buff\n");
4685 	}
4686 
4687 complete:
4688 	if (adapter->fw_version[0] == '\0')
4689 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4690 	complete(&adapter->fw_done);
4691 }
4692 
4693 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4694 {
4695 	struct device *dev = &adapter->vdev->dev;
4696 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4697 	int i;
4698 
4699 	dma_unmap_single(dev, adapter->ip_offload_tok,
4700 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4701 
4702 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4703 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4704 		netdev_dbg(adapter->netdev, "%016lx\n",
4705 			   ((unsigned long *)(buf))[i]);
4706 
4707 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4708 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4709 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4710 		   buf->tcp_ipv4_chksum);
4711 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4712 		   buf->tcp_ipv6_chksum);
4713 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4714 		   buf->udp_ipv4_chksum);
4715 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4716 		   buf->udp_ipv6_chksum);
4717 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4718 		   buf->large_tx_ipv4);
4719 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4720 		   buf->large_tx_ipv6);
4721 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4722 		   buf->large_rx_ipv4);
4723 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4724 		   buf->large_rx_ipv6);
4725 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4726 		   buf->max_ipv4_header_size);
4727 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4728 		   buf->max_ipv6_header_size);
4729 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4730 		   buf->max_tcp_header_size);
4731 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4732 		   buf->max_udp_header_size);
4733 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4734 		   buf->max_large_tx_size);
4735 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4736 		   buf->max_large_rx_size);
4737 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4738 		   buf->ipv6_extension_header);
4739 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4740 		   buf->tcp_pseudosum_req);
4741 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4742 		   buf->num_ipv6_ext_headers);
4743 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4744 		   buf->off_ipv6_ext_headers);
4745 
4746 	send_control_ip_offload(adapter);
4747 }
4748 
4749 static const char *ibmvnic_fw_err_cause(u16 cause)
4750 {
4751 	switch (cause) {
4752 	case ADAPTER_PROBLEM:
4753 		return "adapter problem";
4754 	case BUS_PROBLEM:
4755 		return "bus problem";
4756 	case FW_PROBLEM:
4757 		return "firmware problem";
4758 	case DD_PROBLEM:
4759 		return "device driver problem";
4760 	case EEH_RECOVERY:
4761 		return "EEH recovery";
4762 	case FW_UPDATED:
4763 		return "firmware updated";
4764 	case LOW_MEMORY:
4765 		return "low Memory";
4766 	default:
4767 		return "unknown";
4768 	}
4769 }
4770 
4771 static void handle_error_indication(union ibmvnic_crq *crq,
4772 				    struct ibmvnic_adapter *adapter)
4773 {
4774 	struct device *dev = &adapter->vdev->dev;
4775 	u16 cause;
4776 
4777 	cause = be16_to_cpu(crq->error_indication.error_cause);
4778 
4779 	dev_warn_ratelimited(dev,
4780 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
4781 			     crq->error_indication.flags
4782 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4783 			     ibmvnic_fw_err_cause(cause));
4784 
4785 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4786 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4787 	else
4788 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4789 }
4790 
4791 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4792 				 struct ibmvnic_adapter *adapter)
4793 {
4794 	struct net_device *netdev = adapter->netdev;
4795 	struct device *dev = &adapter->vdev->dev;
4796 	long rc;
4797 
4798 	rc = crq->change_mac_addr_rsp.rc.code;
4799 	if (rc) {
4800 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4801 		goto out;
4802 	}
4803 	/* crq->change_mac_addr.mac_addr is the requested one
4804 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4805 	 */
4806 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
4807 	ether_addr_copy(adapter->mac_addr,
4808 			&crq->change_mac_addr_rsp.mac_addr[0]);
4809 out:
4810 	complete(&adapter->fw_done);
4811 	return rc;
4812 }
4813 
4814 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4815 				   struct ibmvnic_adapter *adapter)
4816 {
4817 	struct device *dev = &adapter->vdev->dev;
4818 	u64 *req_value;
4819 	char *name;
4820 
4821 	atomic_dec(&adapter->running_cap_crqs);
4822 	netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
4823 		   atomic_read(&adapter->running_cap_crqs));
4824 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4825 	case REQ_TX_QUEUES:
4826 		req_value = &adapter->req_tx_queues;
4827 		name = "tx";
4828 		break;
4829 	case REQ_RX_QUEUES:
4830 		req_value = &adapter->req_rx_queues;
4831 		name = "rx";
4832 		break;
4833 	case REQ_RX_ADD_QUEUES:
4834 		req_value = &adapter->req_rx_add_queues;
4835 		name = "rx_add";
4836 		break;
4837 	case REQ_TX_ENTRIES_PER_SUBCRQ:
4838 		req_value = &adapter->req_tx_entries_per_subcrq;
4839 		name = "tx_entries_per_subcrq";
4840 		break;
4841 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4842 		req_value = &adapter->req_rx_add_entries_per_subcrq;
4843 		name = "rx_add_entries_per_subcrq";
4844 		break;
4845 	case REQ_MTU:
4846 		req_value = &adapter->req_mtu;
4847 		name = "mtu";
4848 		break;
4849 	case PROMISC_REQUESTED:
4850 		req_value = &adapter->promisc;
4851 		name = "promisc";
4852 		break;
4853 	default:
4854 		dev_err(dev, "Got invalid cap request rsp %d\n",
4855 			crq->request_capability.capability);
4856 		return;
4857 	}
4858 
4859 	switch (crq->request_capability_rsp.rc.code) {
4860 	case SUCCESS:
4861 		break;
4862 	case PARTIALSUCCESS:
4863 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4864 			 *req_value,
4865 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
4866 			 name);
4867 
4868 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4869 		    REQ_MTU) {
4870 			pr_err("mtu of %llu is not supported. Reverting.\n",
4871 			       *req_value);
4872 			*req_value = adapter->fallback.mtu;
4873 		} else {
4874 			*req_value =
4875 				be64_to_cpu(crq->request_capability_rsp.number);
4876 		}
4877 
4878 		send_request_cap(adapter, 1);
4879 		return;
4880 	default:
4881 		dev_err(dev, "Error %d in request cap rsp\n",
4882 			crq->request_capability_rsp.rc.code);
4883 		return;
4884 	}
4885 
4886 	/* Done receiving requested capabilities, query IP offload support */
4887 	if (atomic_read(&adapter->running_cap_crqs) == 0)
4888 		send_query_ip_offload(adapter);
4889 }
4890 
4891 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4892 			    struct ibmvnic_adapter *adapter)
4893 {
4894 	struct device *dev = &adapter->vdev->dev;
4895 	struct net_device *netdev = adapter->netdev;
4896 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4897 	struct ibmvnic_login_buffer *login = adapter->login_buf;
4898 	u64 *tx_handle_array;
4899 	u64 *rx_handle_array;
4900 	int num_tx_pools;
4901 	int num_rx_pools;
4902 	u64 *size_array;
4903 	int i;
4904 
4905 	/* CHECK: Test/set of login_pending does not need to be atomic
4906 	 * because only ibmvnic_tasklet tests/clears this.
4907 	 */
4908 	if (!adapter->login_pending) {
4909 		netdev_warn(netdev, "Ignoring unexpected login response\n");
4910 		return 0;
4911 	}
4912 	adapter->login_pending = false;
4913 
4914 	dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4915 			 DMA_TO_DEVICE);
4916 	dma_unmap_single(dev, adapter->login_rsp_buf_token,
4917 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4918 
4919 	/* If the number of queues requested can't be allocated by the
4920 	 * server, the login response will return with code 1. We will need
4921 	 * to resend the login buffer with fewer queues requested.
4922 	 */
4923 	if (login_rsp_crq->generic.rc.code) {
4924 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4925 		complete(&adapter->init_done);
4926 		return 0;
4927 	}
4928 
4929 	if (adapter->failover_pending) {
4930 		adapter->init_done_rc = -EAGAIN;
4931 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4932 		complete(&adapter->init_done);
4933 		/* login response buffer will be released on reset */
4934 		return 0;
4935 	}
4936 
4937 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
4938 
4939 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4940 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4941 		netdev_dbg(adapter->netdev, "%016lx\n",
4942 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
4943 	}
4944 
4945 	/* Sanity checks */
4946 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4947 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
4948 	     adapter->req_rx_add_queues !=
4949 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4950 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4951 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4952 		return -EIO;
4953 	}
4954 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4955 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4956 	/* variable buffer sizes are not supported, so just read the
4957 	 * first entry.
4958 	 */
4959 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4960 
4961 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4962 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4963 
4964 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4965 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4966 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4967 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4968 
4969 	for (i = 0; i < num_tx_pools; i++)
4970 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
4971 
4972 	for (i = 0; i < num_rx_pools; i++)
4973 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
4974 
4975 	adapter->num_active_tx_scrqs = num_tx_pools;
4976 	adapter->num_active_rx_scrqs = num_rx_pools;
4977 	release_login_rsp_buffer(adapter);
4978 	release_login_buffer(adapter);
4979 	complete(&adapter->init_done);
4980 
4981 	return 0;
4982 }
4983 
4984 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4985 				     struct ibmvnic_adapter *adapter)
4986 {
4987 	struct device *dev = &adapter->vdev->dev;
4988 	long rc;
4989 
4990 	rc = crq->request_unmap_rsp.rc.code;
4991 	if (rc)
4992 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4993 }
4994 
4995 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4996 				 struct ibmvnic_adapter *adapter)
4997 {
4998 	struct net_device *netdev = adapter->netdev;
4999 	struct device *dev = &adapter->vdev->dev;
5000 	long rc;
5001 
5002 	rc = crq->query_map_rsp.rc.code;
5003 	if (rc) {
5004 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5005 		return;
5006 	}
5007 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5008 		   crq->query_map_rsp.page_size,
5009 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
5010 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
5011 }
5012 
5013 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5014 				 struct ibmvnic_adapter *adapter)
5015 {
5016 	struct net_device *netdev = adapter->netdev;
5017 	struct device *dev = &adapter->vdev->dev;
5018 	long rc;
5019 
5020 	atomic_dec(&adapter->running_cap_crqs);
5021 	netdev_dbg(netdev, "Outstanding queries: %d\n",
5022 		   atomic_read(&adapter->running_cap_crqs));
5023 	rc = crq->query_capability.rc.code;
5024 	if (rc) {
5025 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5026 		goto out;
5027 	}
5028 
5029 	switch (be16_to_cpu(crq->query_capability.capability)) {
5030 	case MIN_TX_QUEUES:
5031 		adapter->min_tx_queues =
5032 		    be64_to_cpu(crq->query_capability.number);
5033 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
5034 			   adapter->min_tx_queues);
5035 		break;
5036 	case MIN_RX_QUEUES:
5037 		adapter->min_rx_queues =
5038 		    be64_to_cpu(crq->query_capability.number);
5039 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
5040 			   adapter->min_rx_queues);
5041 		break;
5042 	case MIN_RX_ADD_QUEUES:
5043 		adapter->min_rx_add_queues =
5044 		    be64_to_cpu(crq->query_capability.number);
5045 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5046 			   adapter->min_rx_add_queues);
5047 		break;
5048 	case MAX_TX_QUEUES:
5049 		adapter->max_tx_queues =
5050 		    be64_to_cpu(crq->query_capability.number);
5051 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5052 			   adapter->max_tx_queues);
5053 		break;
5054 	case MAX_RX_QUEUES:
5055 		adapter->max_rx_queues =
5056 		    be64_to_cpu(crq->query_capability.number);
5057 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5058 			   adapter->max_rx_queues);
5059 		break;
5060 	case MAX_RX_ADD_QUEUES:
5061 		adapter->max_rx_add_queues =
5062 		    be64_to_cpu(crq->query_capability.number);
5063 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5064 			   adapter->max_rx_add_queues);
5065 		break;
5066 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5067 		adapter->min_tx_entries_per_subcrq =
5068 		    be64_to_cpu(crq->query_capability.number);
5069 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5070 			   adapter->min_tx_entries_per_subcrq);
5071 		break;
5072 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5073 		adapter->min_rx_add_entries_per_subcrq =
5074 		    be64_to_cpu(crq->query_capability.number);
5075 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5076 			   adapter->min_rx_add_entries_per_subcrq);
5077 		break;
5078 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5079 		adapter->max_tx_entries_per_subcrq =
5080 		    be64_to_cpu(crq->query_capability.number);
5081 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5082 			   adapter->max_tx_entries_per_subcrq);
5083 		break;
5084 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5085 		adapter->max_rx_add_entries_per_subcrq =
5086 		    be64_to_cpu(crq->query_capability.number);
5087 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5088 			   adapter->max_rx_add_entries_per_subcrq);
5089 		break;
5090 	case TCP_IP_OFFLOAD:
5091 		adapter->tcp_ip_offload =
5092 		    be64_to_cpu(crq->query_capability.number);
5093 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5094 			   adapter->tcp_ip_offload);
5095 		break;
5096 	case PROMISC_SUPPORTED:
5097 		adapter->promisc_supported =
5098 		    be64_to_cpu(crq->query_capability.number);
5099 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5100 			   adapter->promisc_supported);
5101 		break;
5102 	case MIN_MTU:
5103 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5104 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5105 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5106 		break;
5107 	case MAX_MTU:
5108 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5109 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5110 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5111 		break;
5112 	case MAX_MULTICAST_FILTERS:
5113 		adapter->max_multicast_filters =
5114 		    be64_to_cpu(crq->query_capability.number);
5115 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5116 			   adapter->max_multicast_filters);
5117 		break;
5118 	case VLAN_HEADER_INSERTION:
5119 		adapter->vlan_header_insertion =
5120 		    be64_to_cpu(crq->query_capability.number);
5121 		if (adapter->vlan_header_insertion)
5122 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5123 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5124 			   adapter->vlan_header_insertion);
5125 		break;
5126 	case RX_VLAN_HEADER_INSERTION:
5127 		adapter->rx_vlan_header_insertion =
5128 		    be64_to_cpu(crq->query_capability.number);
5129 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5130 			   adapter->rx_vlan_header_insertion);
5131 		break;
5132 	case MAX_TX_SG_ENTRIES:
5133 		adapter->max_tx_sg_entries =
5134 		    be64_to_cpu(crq->query_capability.number);
5135 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5136 			   adapter->max_tx_sg_entries);
5137 		break;
5138 	case RX_SG_SUPPORTED:
5139 		adapter->rx_sg_supported =
5140 		    be64_to_cpu(crq->query_capability.number);
5141 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5142 			   adapter->rx_sg_supported);
5143 		break;
5144 	case OPT_TX_COMP_SUB_QUEUES:
5145 		adapter->opt_tx_comp_sub_queues =
5146 		    be64_to_cpu(crq->query_capability.number);
5147 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5148 			   adapter->opt_tx_comp_sub_queues);
5149 		break;
5150 	case OPT_RX_COMP_QUEUES:
5151 		adapter->opt_rx_comp_queues =
5152 		    be64_to_cpu(crq->query_capability.number);
5153 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5154 			   adapter->opt_rx_comp_queues);
5155 		break;
5156 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5157 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5158 		    be64_to_cpu(crq->query_capability.number);
5159 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5160 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5161 		break;
5162 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5163 		adapter->opt_tx_entries_per_subcrq =
5164 		    be64_to_cpu(crq->query_capability.number);
5165 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5166 			   adapter->opt_tx_entries_per_subcrq);
5167 		break;
5168 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5169 		adapter->opt_rxba_entries_per_subcrq =
5170 		    be64_to_cpu(crq->query_capability.number);
5171 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5172 			   adapter->opt_rxba_entries_per_subcrq);
5173 		break;
5174 	case TX_RX_DESC_REQ:
5175 		adapter->tx_rx_desc_req = crq->query_capability.number;
5176 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5177 			   adapter->tx_rx_desc_req);
5178 		break;
5179 
5180 	default:
5181 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5182 			   crq->query_capability.capability);
5183 	}
5184 
5185 out:
5186 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5187 		send_request_cap(adapter, 0);
5188 }
5189 
5190 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5191 {
5192 	union ibmvnic_crq crq;
5193 	int rc;
5194 
5195 	memset(&crq, 0, sizeof(crq));
5196 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5197 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5198 
5199 	mutex_lock(&adapter->fw_lock);
5200 	adapter->fw_done_rc = 0;
5201 	reinit_completion(&adapter->fw_done);
5202 
5203 	rc = ibmvnic_send_crq(adapter, &crq);
5204 	if (rc) {
5205 		mutex_unlock(&adapter->fw_lock);
5206 		return rc;
5207 	}
5208 
5209 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5210 	if (rc) {
5211 		mutex_unlock(&adapter->fw_lock);
5212 		return rc;
5213 	}
5214 
5215 	mutex_unlock(&adapter->fw_lock);
5216 	return adapter->fw_done_rc ? -EIO : 0;
5217 }
5218 
5219 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5220 				       struct ibmvnic_adapter *adapter)
5221 {
5222 	struct net_device *netdev = adapter->netdev;
5223 	int rc;
5224 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5225 
5226 	rc = crq->query_phys_parms_rsp.rc.code;
5227 	if (rc) {
5228 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5229 		return rc;
5230 	}
5231 	switch (rspeed) {
5232 	case IBMVNIC_10MBPS:
5233 		adapter->speed = SPEED_10;
5234 		break;
5235 	case IBMVNIC_100MBPS:
5236 		adapter->speed = SPEED_100;
5237 		break;
5238 	case IBMVNIC_1GBPS:
5239 		adapter->speed = SPEED_1000;
5240 		break;
5241 	case IBMVNIC_10GBPS:
5242 		adapter->speed = SPEED_10000;
5243 		break;
5244 	case IBMVNIC_25GBPS:
5245 		adapter->speed = SPEED_25000;
5246 		break;
5247 	case IBMVNIC_40GBPS:
5248 		adapter->speed = SPEED_40000;
5249 		break;
5250 	case IBMVNIC_50GBPS:
5251 		adapter->speed = SPEED_50000;
5252 		break;
5253 	case IBMVNIC_100GBPS:
5254 		adapter->speed = SPEED_100000;
5255 		break;
5256 	case IBMVNIC_200GBPS:
5257 		adapter->speed = SPEED_200000;
5258 		break;
5259 	default:
5260 		if (netif_carrier_ok(netdev))
5261 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5262 		adapter->speed = SPEED_UNKNOWN;
5263 	}
5264 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5265 		adapter->duplex = DUPLEX_FULL;
5266 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5267 		adapter->duplex = DUPLEX_HALF;
5268 	else
5269 		adapter->duplex = DUPLEX_UNKNOWN;
5270 
5271 	return rc;
5272 }
5273 
5274 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5275 			       struct ibmvnic_adapter *adapter)
5276 {
5277 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5278 	struct net_device *netdev = adapter->netdev;
5279 	struct device *dev = &adapter->vdev->dev;
5280 	u64 *u64_crq = (u64 *)crq;
5281 	long rc;
5282 
5283 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5284 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5285 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5286 	switch (gen_crq->first) {
5287 	case IBMVNIC_CRQ_INIT_RSP:
5288 		switch (gen_crq->cmd) {
5289 		case IBMVNIC_CRQ_INIT:
5290 			dev_info(dev, "Partner initialized\n");
5291 			adapter->from_passive_init = true;
5292 			/* Discard any stale login responses from prev reset.
5293 			 * CHECK: should we clear even on INIT_COMPLETE?
5294 			 */
5295 			adapter->login_pending = false;
5296 
5297 			if (adapter->state == VNIC_DOWN)
5298 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5299 			else
5300 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5301 
5302 			if (rc && rc != -EBUSY) {
5303 				/* We were unable to schedule the failover
5304 				 * reset either because the adapter was still
5305 				 * probing (eg: during kexec) or we could not
5306 				 * allocate memory. Clear the failover_pending
5307 				 * flag since no one else will. We ignore
5308 				 * EBUSY because it means either FAILOVER reset
5309 				 * is already scheduled or the adapter is
5310 				 * being removed.
5311 				 */
5312 				netdev_err(netdev,
5313 					   "Error %ld scheduling failover reset\n",
5314 					   rc);
5315 				adapter->failover_pending = false;
5316 			}
5317 
5318 			if (!completion_done(&adapter->init_done)) {
5319 				complete(&adapter->init_done);
5320 				if (!adapter->init_done_rc)
5321 					adapter->init_done_rc = -EAGAIN;
5322 			}
5323 
5324 			break;
5325 		case IBMVNIC_CRQ_INIT_COMPLETE:
5326 			dev_info(dev, "Partner initialization complete\n");
5327 			adapter->crq.active = true;
5328 			send_version_xchg(adapter);
5329 			break;
5330 		default:
5331 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5332 		}
5333 		return;
5334 	case IBMVNIC_CRQ_XPORT_EVENT:
5335 		netif_carrier_off(netdev);
5336 		adapter->crq.active = false;
5337 		/* terminate any thread waiting for a response
5338 		 * from the device
5339 		 */
5340 		if (!completion_done(&adapter->fw_done)) {
5341 			adapter->fw_done_rc = -EIO;
5342 			complete(&adapter->fw_done);
5343 		}
5344 		if (!completion_done(&adapter->stats_done))
5345 			complete(&adapter->stats_done);
5346 		if (test_bit(0, &adapter->resetting))
5347 			adapter->force_reset_recovery = true;
5348 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5349 			dev_info(dev, "Migrated, re-enabling adapter\n");
5350 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5351 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5352 			dev_info(dev, "Backing device failover detected\n");
5353 			adapter->failover_pending = true;
5354 		} else {
5355 			/* The adapter lost the connection */
5356 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5357 				gen_crq->cmd);
5358 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5359 		}
5360 		return;
5361 	case IBMVNIC_CRQ_CMD_RSP:
5362 		break;
5363 	default:
5364 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
5365 			gen_crq->first);
5366 		return;
5367 	}
5368 
5369 	switch (gen_crq->cmd) {
5370 	case VERSION_EXCHANGE_RSP:
5371 		rc = crq->version_exchange_rsp.rc.code;
5372 		if (rc) {
5373 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5374 			break;
5375 		}
5376 		ibmvnic_version =
5377 			    be16_to_cpu(crq->version_exchange_rsp.version);
5378 		dev_info(dev, "Partner protocol version is %d\n",
5379 			 ibmvnic_version);
5380 		send_query_cap(adapter);
5381 		break;
5382 	case QUERY_CAPABILITY_RSP:
5383 		handle_query_cap_rsp(crq, adapter);
5384 		break;
5385 	case QUERY_MAP_RSP:
5386 		handle_query_map_rsp(crq, adapter);
5387 		break;
5388 	case REQUEST_MAP_RSP:
5389 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5390 		complete(&adapter->fw_done);
5391 		break;
5392 	case REQUEST_UNMAP_RSP:
5393 		handle_request_unmap_rsp(crq, adapter);
5394 		break;
5395 	case REQUEST_CAPABILITY_RSP:
5396 		handle_request_cap_rsp(crq, adapter);
5397 		break;
5398 	case LOGIN_RSP:
5399 		netdev_dbg(netdev, "Got Login Response\n");
5400 		handle_login_rsp(crq, adapter);
5401 		break;
5402 	case LOGICAL_LINK_STATE_RSP:
5403 		netdev_dbg(netdev,
5404 			   "Got Logical Link State Response, state: %d rc: %d\n",
5405 			   crq->logical_link_state_rsp.link_state,
5406 			   crq->logical_link_state_rsp.rc.code);
5407 		adapter->logical_link_state =
5408 		    crq->logical_link_state_rsp.link_state;
5409 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5410 		complete(&adapter->init_done);
5411 		break;
5412 	case LINK_STATE_INDICATION:
5413 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
5414 		adapter->phys_link_state =
5415 		    crq->link_state_indication.phys_link_state;
5416 		adapter->logical_link_state =
5417 		    crq->link_state_indication.logical_link_state;
5418 		if (adapter->phys_link_state && adapter->logical_link_state)
5419 			netif_carrier_on(netdev);
5420 		else
5421 			netif_carrier_off(netdev);
5422 		break;
5423 	case CHANGE_MAC_ADDR_RSP:
5424 		netdev_dbg(netdev, "Got MAC address change Response\n");
5425 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5426 		break;
5427 	case ERROR_INDICATION:
5428 		netdev_dbg(netdev, "Got Error Indication\n");
5429 		handle_error_indication(crq, adapter);
5430 		break;
5431 	case REQUEST_STATISTICS_RSP:
5432 		netdev_dbg(netdev, "Got Statistics Response\n");
5433 		complete(&adapter->stats_done);
5434 		break;
5435 	case QUERY_IP_OFFLOAD_RSP:
5436 		netdev_dbg(netdev, "Got Query IP offload Response\n");
5437 		handle_query_ip_offload_rsp(adapter);
5438 		break;
5439 	case MULTICAST_CTRL_RSP:
5440 		netdev_dbg(netdev, "Got multicast control Response\n");
5441 		break;
5442 	case CONTROL_IP_OFFLOAD_RSP:
5443 		netdev_dbg(netdev, "Got Control IP offload Response\n");
5444 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5445 				 sizeof(adapter->ip_offload_ctrl),
5446 				 DMA_TO_DEVICE);
5447 		complete(&adapter->init_done);
5448 		break;
5449 	case COLLECT_FW_TRACE_RSP:
5450 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5451 		complete(&adapter->fw_done);
5452 		break;
5453 	case GET_VPD_SIZE_RSP:
5454 		handle_vpd_size_rsp(crq, adapter);
5455 		break;
5456 	case GET_VPD_RSP:
5457 		handle_vpd_rsp(crq, adapter);
5458 		break;
5459 	case QUERY_PHYS_PARMS_RSP:
5460 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5461 		complete(&adapter->fw_done);
5462 		break;
5463 	default:
5464 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5465 			   gen_crq->cmd);
5466 	}
5467 }
5468 
5469 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5470 {
5471 	struct ibmvnic_adapter *adapter = instance;
5472 
5473 	tasklet_schedule(&adapter->tasklet);
5474 	return IRQ_HANDLED;
5475 }
5476 
5477 static void ibmvnic_tasklet(struct tasklet_struct *t)
5478 {
5479 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5480 	struct ibmvnic_crq_queue *queue = &adapter->crq;
5481 	union ibmvnic_crq *crq;
5482 	unsigned long flags;
5483 
5484 	spin_lock_irqsave(&queue->lock, flags);
5485 
5486 	/* Pull all the valid messages off the CRQ */
5487 	while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5488 		/* This barrier makes sure ibmvnic_next_crq()'s
5489 		 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5490 		 * before ibmvnic_handle_crq()'s
5491 		 * switch(gen_crq->first) and switch(gen_crq->cmd).
5492 		 */
5493 		dma_rmb();
5494 		ibmvnic_handle_crq(crq, adapter);
5495 		crq->generic.first = 0;
5496 	}
5497 
5498 	spin_unlock_irqrestore(&queue->lock, flags);
5499 }
5500 
5501 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5502 {
5503 	struct vio_dev *vdev = adapter->vdev;
5504 	int rc;
5505 
5506 	do {
5507 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5508 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5509 
5510 	if (rc)
5511 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5512 
5513 	return rc;
5514 }
5515 
5516 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5517 {
5518 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5519 	struct device *dev = &adapter->vdev->dev;
5520 	struct vio_dev *vdev = adapter->vdev;
5521 	int rc;
5522 
5523 	/* Close the CRQ */
5524 	do {
5525 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5526 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5527 
5528 	/* Clean out the queue */
5529 	if (!crq->msgs)
5530 		return -EINVAL;
5531 
5532 	memset(crq->msgs, 0, PAGE_SIZE);
5533 	crq->cur = 0;
5534 	crq->active = false;
5535 
5536 	/* And re-open it again */
5537 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5538 				crq->msg_token, PAGE_SIZE);
5539 
5540 	if (rc == H_CLOSED)
5541 		/* Adapter is good, but other end is not ready */
5542 		dev_warn(dev, "Partner adapter not ready\n");
5543 	else if (rc != 0)
5544 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5545 
5546 	return rc;
5547 }
5548 
5549 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5550 {
5551 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5552 	struct vio_dev *vdev = adapter->vdev;
5553 	long rc;
5554 
5555 	if (!crq->msgs)
5556 		return;
5557 
5558 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5559 	free_irq(vdev->irq, adapter);
5560 	tasklet_kill(&adapter->tasklet);
5561 	do {
5562 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5563 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5564 
5565 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5566 			 DMA_BIDIRECTIONAL);
5567 	free_page((unsigned long)crq->msgs);
5568 	crq->msgs = NULL;
5569 	crq->active = false;
5570 }
5571 
5572 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5573 {
5574 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5575 	struct device *dev = &adapter->vdev->dev;
5576 	struct vio_dev *vdev = adapter->vdev;
5577 	int rc, retrc = -ENOMEM;
5578 
5579 	if (crq->msgs)
5580 		return 0;
5581 
5582 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5583 	/* Should we allocate more than one page? */
5584 
5585 	if (!crq->msgs)
5586 		return -ENOMEM;
5587 
5588 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5589 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5590 					DMA_BIDIRECTIONAL);
5591 	if (dma_mapping_error(dev, crq->msg_token))
5592 		goto map_failed;
5593 
5594 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5595 				crq->msg_token, PAGE_SIZE);
5596 
5597 	if (rc == H_RESOURCE)
5598 		/* maybe kexecing and resource is busy. try a reset */
5599 		rc = ibmvnic_reset_crq(adapter);
5600 	retrc = rc;
5601 
5602 	if (rc == H_CLOSED) {
5603 		dev_warn(dev, "Partner adapter not ready\n");
5604 	} else if (rc) {
5605 		dev_warn(dev, "Error %d opening adapter\n", rc);
5606 		goto reg_crq_failed;
5607 	}
5608 
5609 	retrc = 0;
5610 
5611 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5612 
5613 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5614 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5615 		 adapter->vdev->unit_address);
5616 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5617 	if (rc) {
5618 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5619 			vdev->irq, rc);
5620 		goto req_irq_failed;
5621 	}
5622 
5623 	rc = vio_enable_interrupts(vdev);
5624 	if (rc) {
5625 		dev_err(dev, "Error %d enabling interrupts\n", rc);
5626 		goto req_irq_failed;
5627 	}
5628 
5629 	crq->cur = 0;
5630 	spin_lock_init(&crq->lock);
5631 
5632 	/* process any CRQs that were queued before we enabled interrupts */
5633 	tasklet_schedule(&adapter->tasklet);
5634 
5635 	return retrc;
5636 
5637 req_irq_failed:
5638 	tasklet_kill(&adapter->tasklet);
5639 	do {
5640 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5641 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5642 reg_crq_failed:
5643 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5644 map_failed:
5645 	free_page((unsigned long)crq->msgs);
5646 	crq->msgs = NULL;
5647 	return retrc;
5648 }
5649 
5650 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5651 {
5652 	struct device *dev = &adapter->vdev->dev;
5653 	unsigned long timeout = msecs_to_jiffies(20000);
5654 	u64 old_num_rx_queues = adapter->req_rx_queues;
5655 	u64 old_num_tx_queues = adapter->req_tx_queues;
5656 	int rc;
5657 
5658 	adapter->from_passive_init = false;
5659 
5660 	if (reset)
5661 		reinit_completion(&adapter->init_done);
5662 
5663 	adapter->init_done_rc = 0;
5664 	rc = ibmvnic_send_crq_init(adapter);
5665 	if (rc) {
5666 		dev_err(dev, "Send crq init failed with error %d\n", rc);
5667 		return rc;
5668 	}
5669 
5670 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5671 		dev_err(dev, "Initialization sequence timed out\n");
5672 		return -ETIMEDOUT;
5673 	}
5674 
5675 	if (adapter->init_done_rc) {
5676 		release_crq_queue(adapter);
5677 		return adapter->init_done_rc;
5678 	}
5679 
5680 	if (adapter->from_passive_init) {
5681 		adapter->state = VNIC_OPEN;
5682 		adapter->from_passive_init = false;
5683 		return -EINVAL;
5684 	}
5685 
5686 	if (reset &&
5687 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5688 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
5689 		if (adapter->req_rx_queues != old_num_rx_queues ||
5690 		    adapter->req_tx_queues != old_num_tx_queues) {
5691 			release_sub_crqs(adapter, 0);
5692 			rc = init_sub_crqs(adapter);
5693 		} else {
5694 			rc = reset_sub_crq_queues(adapter);
5695 		}
5696 	} else {
5697 		rc = init_sub_crqs(adapter);
5698 	}
5699 
5700 	if (rc) {
5701 		dev_err(dev, "Initialization of sub crqs failed\n");
5702 		release_crq_queue(adapter);
5703 		return rc;
5704 	}
5705 
5706 	rc = init_sub_crq_irqs(adapter);
5707 	if (rc) {
5708 		dev_err(dev, "Failed to initialize sub crq irqs\n");
5709 		release_crq_queue(adapter);
5710 	}
5711 
5712 	return rc;
5713 }
5714 
5715 static struct device_attribute dev_attr_failover;
5716 
5717 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5718 {
5719 	struct ibmvnic_adapter *adapter;
5720 	struct net_device *netdev;
5721 	unsigned char *mac_addr_p;
5722 	bool init_success;
5723 	int rc;
5724 
5725 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5726 		dev->unit_address);
5727 
5728 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5729 							VETH_MAC_ADDR, NULL);
5730 	if (!mac_addr_p) {
5731 		dev_err(&dev->dev,
5732 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5733 			__FILE__, __LINE__);
5734 		return 0;
5735 	}
5736 
5737 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5738 				   IBMVNIC_MAX_QUEUES);
5739 	if (!netdev)
5740 		return -ENOMEM;
5741 
5742 	adapter = netdev_priv(netdev);
5743 	adapter->state = VNIC_PROBING;
5744 	dev_set_drvdata(&dev->dev, netdev);
5745 	adapter->vdev = dev;
5746 	adapter->netdev = netdev;
5747 	adapter->login_pending = false;
5748 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
5749 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
5750 	bitmap_set(adapter->map_ids, 0, 1);
5751 
5752 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
5753 	eth_hw_addr_set(netdev, adapter->mac_addr);
5754 	netdev->irq = dev->irq;
5755 	netdev->netdev_ops = &ibmvnic_netdev_ops;
5756 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5757 	SET_NETDEV_DEV(netdev, &dev->dev);
5758 
5759 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5760 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5761 			  __ibmvnic_delayed_reset);
5762 	INIT_LIST_HEAD(&adapter->rwi_list);
5763 	spin_lock_init(&adapter->rwi_lock);
5764 	spin_lock_init(&adapter->state_lock);
5765 	mutex_init(&adapter->fw_lock);
5766 	init_completion(&adapter->init_done);
5767 	init_completion(&adapter->fw_done);
5768 	init_completion(&adapter->reset_done);
5769 	init_completion(&adapter->stats_done);
5770 	clear_bit(0, &adapter->resetting);
5771 	adapter->prev_rx_buf_sz = 0;
5772 	adapter->prev_mtu = 0;
5773 
5774 	init_success = false;
5775 	do {
5776 		rc = init_crq_queue(adapter);
5777 		if (rc) {
5778 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5779 				rc);
5780 			goto ibmvnic_init_fail;
5781 		}
5782 
5783 		rc = ibmvnic_reset_init(adapter, false);
5784 	} while (rc == -EAGAIN);
5785 
5786 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
5787 	 * partner is not ready. CRQ is not active. When the partner becomes
5788 	 * ready, we will do the passive init reset.
5789 	 */
5790 
5791 	if (!rc)
5792 		init_success = true;
5793 
5794 	rc = init_stats_buffers(adapter);
5795 	if (rc)
5796 		goto ibmvnic_init_fail;
5797 
5798 	rc = init_stats_token(adapter);
5799 	if (rc)
5800 		goto ibmvnic_stats_fail;
5801 
5802 	rc = device_create_file(&dev->dev, &dev_attr_failover);
5803 	if (rc)
5804 		goto ibmvnic_dev_file_err;
5805 
5806 	netif_carrier_off(netdev);
5807 	rc = register_netdev(netdev);
5808 	if (rc) {
5809 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5810 		goto ibmvnic_register_fail;
5811 	}
5812 	dev_info(&dev->dev, "ibmvnic registered\n");
5813 
5814 	if (init_success) {
5815 		adapter->state = VNIC_PROBED;
5816 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
5817 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5818 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5819 	} else {
5820 		adapter->state = VNIC_DOWN;
5821 	}
5822 
5823 	adapter->wait_for_reset = false;
5824 	adapter->last_reset_time = jiffies;
5825 	return 0;
5826 
5827 ibmvnic_register_fail:
5828 	device_remove_file(&dev->dev, &dev_attr_failover);
5829 
5830 ibmvnic_dev_file_err:
5831 	release_stats_token(adapter);
5832 
5833 ibmvnic_stats_fail:
5834 	release_stats_buffers(adapter);
5835 
5836 ibmvnic_init_fail:
5837 	release_sub_crqs(adapter, 1);
5838 	release_crq_queue(adapter);
5839 	mutex_destroy(&adapter->fw_lock);
5840 	free_netdev(netdev);
5841 
5842 	return rc;
5843 }
5844 
5845 static void ibmvnic_remove(struct vio_dev *dev)
5846 {
5847 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
5848 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5849 	unsigned long flags;
5850 
5851 	spin_lock_irqsave(&adapter->state_lock, flags);
5852 
5853 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
5854 	 * finish. Then, set the state to REMOVING to prevent it from
5855 	 * scheduling any more work and to have reset functions ignore
5856 	 * any resets that have already been scheduled. Drop the lock
5857 	 * after setting state, so __ibmvnic_reset() which is called
5858 	 * from the flush_work() below, can make progress.
5859 	 */
5860 	spin_lock(&adapter->rwi_lock);
5861 	adapter->state = VNIC_REMOVING;
5862 	spin_unlock(&adapter->rwi_lock);
5863 
5864 	spin_unlock_irqrestore(&adapter->state_lock, flags);
5865 
5866 	flush_work(&adapter->ibmvnic_reset);
5867 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5868 
5869 	rtnl_lock();
5870 	unregister_netdevice(netdev);
5871 
5872 	release_resources(adapter);
5873 	release_rx_pools(adapter);
5874 	release_tx_pools(adapter);
5875 	release_sub_crqs(adapter, 1);
5876 	release_crq_queue(adapter);
5877 
5878 	release_stats_token(adapter);
5879 	release_stats_buffers(adapter);
5880 
5881 	adapter->state = VNIC_REMOVED;
5882 
5883 	rtnl_unlock();
5884 	mutex_destroy(&adapter->fw_lock);
5885 	device_remove_file(&dev->dev, &dev_attr_failover);
5886 	free_netdev(netdev);
5887 	dev_set_drvdata(&dev->dev, NULL);
5888 }
5889 
5890 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5891 			      const char *buf, size_t count)
5892 {
5893 	struct net_device *netdev = dev_get_drvdata(dev);
5894 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5895 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5896 	__be64 session_token;
5897 	long rc;
5898 
5899 	if (!sysfs_streq(buf, "1"))
5900 		return -EINVAL;
5901 
5902 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5903 			 H_GET_SESSION_TOKEN, 0, 0, 0);
5904 	if (rc) {
5905 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5906 			   rc);
5907 		goto last_resort;
5908 	}
5909 
5910 	session_token = (__be64)retbuf[0];
5911 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5912 		   be64_to_cpu(session_token));
5913 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5914 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
5915 	if (rc)
5916 		netdev_err(netdev,
5917 			   "H_VIOCTL initiated failover failed, rc %ld\n",
5918 			   rc);
5919 
5920 last_resort:
5921 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
5922 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5923 
5924 	return count;
5925 }
5926 static DEVICE_ATTR_WO(failover);
5927 
5928 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5929 {
5930 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5931 	struct ibmvnic_adapter *adapter;
5932 	struct iommu_table *tbl;
5933 	unsigned long ret = 0;
5934 	int i;
5935 
5936 	tbl = get_iommu_table_base(&vdev->dev);
5937 
5938 	/* netdev inits at probe time along with the structures we need below*/
5939 	if (!netdev)
5940 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5941 
5942 	adapter = netdev_priv(netdev);
5943 
5944 	ret += PAGE_SIZE; /* the crq message queue */
5945 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5946 
5947 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5948 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
5949 
5950 	for (i = 0; i < adapter->num_active_rx_pools; i++)
5951 		ret += adapter->rx_pool[i].size *
5952 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5953 
5954 	return ret;
5955 }
5956 
5957 static int ibmvnic_resume(struct device *dev)
5958 {
5959 	struct net_device *netdev = dev_get_drvdata(dev);
5960 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5961 
5962 	if (adapter->state != VNIC_OPEN)
5963 		return 0;
5964 
5965 	tasklet_schedule(&adapter->tasklet);
5966 
5967 	return 0;
5968 }
5969 
5970 static const struct vio_device_id ibmvnic_device_table[] = {
5971 	{"network", "IBM,vnic"},
5972 	{"", "" }
5973 };
5974 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5975 
5976 static const struct dev_pm_ops ibmvnic_pm_ops = {
5977 	.resume = ibmvnic_resume
5978 };
5979 
5980 static struct vio_driver ibmvnic_driver = {
5981 	.id_table       = ibmvnic_device_table,
5982 	.probe          = ibmvnic_probe,
5983 	.remove         = ibmvnic_remove,
5984 	.get_desired_dma = ibmvnic_get_desired_dma,
5985 	.name		= ibmvnic_driver_name,
5986 	.pm		= &ibmvnic_pm_ops,
5987 };
5988 
5989 /* module functions */
5990 static int __init ibmvnic_module_init(void)
5991 {
5992 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5993 		IBMVNIC_DRIVER_VERSION);
5994 
5995 	return vio_register_driver(&ibmvnic_driver);
5996 }
5997 
5998 static void __exit ibmvnic_module_exit(void)
5999 {
6000 	vio_unregister_driver(&ibmvnic_driver);
6001 }
6002 
6003 module_init(ibmvnic_module_init);
6004 module_exit(ibmvnic_module_exit);
6005